Signals | Commerce in the age of generative AI

Bias in generative AI can originate in inadequate training data that encodes societal biases, sometimes unwittingly amplified by developers. Image-generating tools could reinforce archaic stereotypes about race, gender and professions, simply because they’re trained on old data sets. Efforts to fight bias should include embedding diverse viewpoints, regular audits and collaborative work among AI researchers, ethicists, policymakers and communities impacted by this issue.

Cautionary calls

Legislators, government officials, researchers, private sector decision-makers and employees, and other stakeholders should adhere to best practices in their AI efforts. All of their actions should be consistent with the following values:

In 2023, generative AI has established a distinct role in the technology dialogue. With OpenAI's ChatGPT, individuals with no technical expertise suddenly found themselves exposed to the remarkable power of this technology. While AI has been around for decades, three primary factors have contributed to generative AI's rapid advancement:

To learn more about changing ideas of value and money and their implications for businesses, individuals and society at large, please look out for the Q3 2023 issue of Mastercard’s thought leadership publication Signals, which will explore the topic of reimagined money.

Read more about Mastercard's activity in this space

Produced by Mastercard Foundry

View sources

Hide sources

Retail and e-commerce entities can employ generative AI to optimize supply chains, bolster marketing efforts through hyper-personalization, improve measurement and accelerate the speed to market for new products and services.

Over the next two to three years, generative AI will power hundreds of capabilities across business and consumer applications. Here are a just a few of the many examples:

Signals

Q3 2023

Emerging use cases

Large enterprises employ thousands of workers across dozens of disciplines and hundreds of products and services. Enterprise learning systems can therefore be complex, encompassing team meetings, town halls, employee training sessions, internal communications, and quarterly reviews, to name a few. Generative AI has the potential to make corporate collaborations much more agile. Via machine learning capabilities, AI tools can facilitate the horizontal distribution of information in near real-time — imagine knowledge bots instantaneously offering insights in strategy sessions. The corporate landscape will take on a new dynamism as employees operate with increased speed and flexibility and processes become streamlined. For example, enterprise search engines have traditionally identified unstructured information stored within documents, wikis, blogs, and intranets by indexing the body of the content or the metadata — but generative AI tools can provide answers, not just a list of documents. Companies are now developing workplace-specific search tools that use knowledge models for more relevant results. Integrated solutions using generative AI can assist employees in accessing the necessary information they need to perform their jobs effectively, no matter where that knowledge resides.

Distributing knowledge and insights

Data integrity remains a significant challenge, because LLMs and chatbots can provide false information. A knowledge distribution model or system will be functional, correct, and relevant only when it’s adequately connected to the right data, with proper governance and permissions embedded. Improvements must address "hallucinations," nonsensical advice and fabricated details that gen AI models sometimes produce. Integrating and curating closed data sets can enhance reliability, but completely automating knowledge-sharing and performing it without human verification can exacerbate errors.

Simplifying wealth management

Today's financial ecosystem is marked by its complexity, requiring interactions between institutions — including banks, insurance companies, investment firms and governmental entities — for taxation and property registration purposes. Generative AI, in synergy with informed data consent protocols, could declutter and streamline these processes, effectively acting as a personal wealth manager with an encompassing view of an individual’s financial life. It can seamlessly integrate bank accounts, investment portfolios and small business accounts, and even interact with governmental revenue offices through open banking data methods. Gen AI also can bolster financial literacy and provide insights into best practices, essentially serving as a family wealth management office. Envision a scenario where an AI-enhanced conversational engine helps formulate college savings plans, procure loans and implement financial strategies, including sophisticated ones that are typically beyond an individual investor. Generative AI’s personalized recommendations and expertise could empower people to navigate their financial lives more adeptly.

The democratization of high-caliber financial services through generative AI could foster economic equity for some, as retail investors equipped with AI-enhanced tools could potentially rival the performances of professionals in legacy financial offices. But there also is the risk this could widen the chasm between an increasingly well-equipped middle class and the unbanked and underbanked population — those approximately 1.4 billion individuals globally who lack rudimentary banking services, let alone sophisticated trading platforms.

Morgan Stanley Wealth Management exemplifies the banking sector's early adoption of AI. Through strategic collaboration with OpenAI, it is capitalizing on AI technology and its own intellectual assets to give financial advisors real-time insights and content. Key initiatives include Next Best Action, an AI engine that empowers financial advisors to create customized communications for clients and prospects, and Genome an internal technological venture that leverages data analytics and fintech alliances to tailor offerings for clientele across the bank's workspace and E-Trade channels.

Enabling new capabilities

Small businesses have evolved over the last several years to become increasingly digital. At the same time, the gig and creator economies have grown, resulting in more companies being formed as one-person enterprises. These entrepreneurs grapple with resource limitations and insufficient bandwidth, often playing multiple roles within their businesses. Generative AI can be an invaluable tool for supporting solopreneurs and small businesses by adding AI knowledge workers to the team. Business owners can employ AI as digital CFOs offering financial management and digital CMOs orchestrating marketing campaigns. Microsoft has rolled out AI copilots tailor-made for a broad spectrum of professionals, including financial analysts, commercial real estate developers, attorneys and others. The near-term use cases include automating content creation, significantly reducing the time and effort required to generate marketing material or online content. As generative AI improves, new opportunities will likely open for creators to quickly launch new products, services, and companies. For those lacking technical acumen, generative AI offers coding capabilities that let them build prototype apps and translate their visions into tangible solutions. In the future, coding skills may not be a prerequisite for video game creation and technical production prowess may not be a requirement for aspiring filmmakers. Entrepreneurs can capitalize on potent new tools to build new ventures faster and scale their offerings via automated sales and revenue growth bots.

AI-powered personal shoppers

E-commerce leaders like Amazon and Alibaba offer more choices, helpful customer testimonials, and price comparisons. However, the sheer volume of options can inundate consumers and lead to indecisiveness. Moreover, the quest for specific items may require scouring a variety of channels, making the online shopping experience more complex. Enter AI-powered personal shopping consultants poised to redefine consumer convenience. Picture a scenario where a virtual shopping expert scans multiple channels, weeds out products with bad reviews, pinpoint the most cost-effective options, and retrieves the exact items you seek. Armed with an intricate understanding of your preferences, these automated shopping consultants could outperform human advisors. It can provide personalized product recommendations to shoppers by analyzing customer data and generating contextually relevant product recommendations and offers. By assimilating body measurement and biometric data, AI assistants can curate apparel that fits flawlessly. Furthermore, merchants can use the technology to optimize product descriptions and e-commerce platforms so shoppers get a hyper-personalized storefront. With AI, the e-commerce experience transforms from essentially a search exercise into an intelligent, personalized experience.

Socioeconomic divides

Platform integrations

Information validity

The re-emergence of the agent

Organizing a trip can often feel like assembling an elaborate jigsaw puzzle, requiring travelers to piece together myriad components across many time zones and currencies. AI-facilitated automation, however, can simplify the process. Through the integration of a conversational interface with an array of travel platforms — those of airlines, hospitality providers and transportation services — planning a trip could be reduced to a brief command such as, "Organize a vacation to an Italian coastal village like Positano, but with fewer crowds and within my budget." Moreover, AI is equipped to deliver astute price predictions, giving travelers critical insights into airline fare trends, thereby ensuring they’re offered optimal deals. Imagine employing a voice interface on a platform like Expedia: Rather than being inundated with options, you’ll receive AI-crafted itineraries with confirmed accommodations, transportation bookings and dining reservations tailored to your preferences. Generative AI transcends the limitations of logistics and planning; it can synthesize immersive videos that offer glimpses into travelers' impending adventures. Utilizing cutting-edge technologies like Apple's new Vision Pro spatial computer, individuals can conduct virtual surveys of their journeys, ensuring thorough preparation and heightening anticipation.

In April 2023, China drafted rules that required companies to register generative AI products and submit them for security assessments. More recently, it has adopted a softer approach designed to encourage Chinese companies to compete globally.

The following trends could define generative AI’s development in the next five to seven years:

AI models require vast amounts of training data, but high-value data is often confined within proprietary systems. Entities with critical data, such as large banks and tech companies, will be at the forefront of development and companies that excel in data security will thrive.

At Mastercard, AI models already serve as linchpins to our solutions, safeguarding more than 125 billion transactions on our network every year. Employing hundreds of data scientists, AI technologists, and a growing team of AI governance experts, we’re committed to developing practical AI solutions that integrate privacy and ethics by design. Across our capabilities — data intelligence, open banking, identity, fraud protection and cybersecurity — Mastercard ensures trust is at the forefront, and AI is used responsibly and ethically. We expect generative AI to evolve considerably over the coming years. Even in its nascent form, this powerful technology promises incredible opportunities when combined with human oversight. Safe usage depends on our collective actions as a global community. By embracing generative AI while addressing its risks and challenges, we can step into the future of commerce and make sure it influences the world for the greater good.

Money will include tokenized assets and other new forms of value.

Complex, conditional commercial payments will be automated to speed commerce.

Programmable payments

Next-gen e-wallets will manage our identities, assets, payments and more.

Ubiquitous wallets

Our assets will be accessible in any environment.

Connected finance

Payments will break through today’s geographic and digital boundaries.

Borderless rails

Next-gen points of interaction will drive new ways for consumers to pay.

Unleashing acceptance

New financing solutions will empower underbanked people and communities.

Inclusive credit

Consumers will increasingly spend with companies that align with their values.

Conscious consumerism

Trust will become a critical point of differentiation for companies.

Embedded trust

Money will include tokenized assets and other new forms of value.

A tokenized world

Complex, conditional commercial payments will be automated to speed commerce.

Programmable payments

Next-gen e-wallets will manage our identities, assets, payments and more.

Ubiquitous wallets

Our assets will be accessible in any environment.

Connected finance

Payments will break through today’s geographic and digital boundaries.

Borderless rails

Next-gen points of interaction will drive new ways for consumers to pay.

Unleashing acceptance

As business leaders evaluate this technology's strategic applications, increased accessibility to data will boost generative AI's potential and let organizations deploy solutions within their own controlled environments.

Larry Page, Co-founder of Google

In early 2023, OpenAI introduced plug-ins, allowing companies like Expedia, Instacart and Klarna to expose their APIs to a conversational interface without the need for users to program the interface directly. This critical development helped transform generative AI into a practical tool that enhances consumer experiences. ChatGPT's plug-in feature lets website users search for up-to-date information and interact with online shopping carts, improving recommendations and streamlining the online journey.

ChatGPT has primarily been available as a hosted model, which means OpenAI stores and has access to all the data people input into its chatbot. That type of system is problematic for companies that typically need to control their non-public, proprietary and customer data. Open-source technologies, like Meta's LLaMa, in which individual companies can manage the storage and access of their data, empower organizations to use generative AI safely with their data without disclosing it publicly.

Generative AI models can be made more relevant for vertical use cases or personalized customer applications when they are trained on proprietary data. Models that can access and learn from specific data, such as transaction history, can provide better banking interactions. Enterprises looking to build bespoke gen AI applications can leverage tokenization to mask the underlying data and facilitate the safe sharing of private information. For example, single-use tokens can make sharing data possible in payments when using gen AI during a checkout journey.

Open banking, which allows users to share their banking data so they can access fintech and other services, is another enabler. Open banking platforms' existing privacy and security frameworks will let consumers control the use of their data by these AI models. Through open banking, generative AI can access a broader dataset and consequently create more sophisticated models in specific verticals.

A more efficient attention economy

What traditional search engines are to information, generative AI is to answers — perhaps ending the need to wade through search results. Taken a step further: Search engines put pages in front of you like a research librarian stacks books on a counter, while generative AI models act as study partner, continually refining answers in a back-and-forth conversation. AI-powered semantic search — which considers meaning, intention and context — also holds promise.

Commerce in the age of generative AI

Generative AI will improve content quality and efficiency, paramount for the gaming, film and music industries. It can facilitate immersive commerce experiences by creating hyper-realistic simulations and virtual environments.

GPT-4 outperforms ChatGPT by scoring in higher approximate percentiles among test-takers

Uniform bar exam

Biology olympiad

10th

90th

31st

99th

Initially focused on open-source research, OpenAI morphed into a for-profit entity offering access to advanced language models such as GPT4 and DALL-E.

We are now encountering another inflection point in the Next Economy: the democratization of generative AI, a technology garnering considerable attention for its capacity to emulate human expression and produce human-like content. The rise of platforms such as ChatGPT has triggered this surge in interest. Even early on, generative AI promises to yield remarkable results, from transforming consumer experiences through intelligent shopping to enhancing communications via automated email tools. For businesses, generative AI has the potential to improve customer engagement through hyper-realistic conversational agents and tailored advertising and to usher in efficiencies by automating and optimizing workflows. It also could support software development, improve knowledge management and much more. In this issue of Signals, we cover the latest developments in generative AI, delve into what it makes possible, weigh the opportunities it presents, and assess its challenges. We aim to separate signal from noise and explore how generative AI will advance the Next Economy.

The landscape of commerce is undergoing seismic shifts. On the horizon is the Next Economy, Mastercard's vision for a future of commerce underpinned by innovations that will let us reimagine money and value exchange, promote intelligent experiences, and entrench principles of inclusion and sustainability into the way we develop products and services. Banks, merchants and digital players are poised to be at the forefront of this revolution, putting to work emerging technologies such as tokenization, cloud computing and advanced networking.

Decoding gen AI

The sheer volume of data needed to fuel AI learning processes is readily available due to the exponential growth of both structured and unstructured data found online and within enterprises.

Advancements in central and graphics processing units have amplified their capability, providing the necessary computational muscle for AI mechanisms to generate swift outputs.

The maturation of deep learning algorithms, particularly artificial neural networks, has resulted in their ability to mimic more intricate thought processes.

“Artificial intelligence would be the ultimate version of Google. The ultimate search engine that would understand everything on the web. It would understand exactly what you wanted, and it would give you the right thing.”

Plug-ins

Open-source tech

Proprietary data

Connected data

Transforming commerce

Financial institutions will leverage generative AI for personalized banking services, fraud detection and regulatory compliance, among other uses. Gen AI is instrumental in risk mitigation, as it can generate synthetic data sets to test the efficacy of fraud detection algorithms.

CHALLENGES

SynthAI

EARLY SIGNALS

Training models on an enterprise's internal data is central to knowledge sharing. Venture capital firm A16Z has posited that the future of AI in the workplace may primarily involve something other than LLMs like ChatGPT. Instead, they envision a shift towards more specialized AI models, termed SynthAI, tailored to address specific business needs. These models leverage proprietary datasets fine-tuned for tasks like resolving support issues and summarizing market research.

Challenges

Early Signals

Fiscal and technical constraints

The data- and computing-intensive nature of generative AI renders it costly. According to one estimate, running ChatGPT costs OpenAI up to $700,000 daily in operating expenses. These costs will trickle down to users. There is also the constraint involved in connecting AI models to data. Deploying an AI-powered “virtual CFO” may require integrating the generative AI model with internal data across several applications, potentially a big lift for a small organization.

Small business owners can use AI to amalgamate and automate an array of tools and platforms. For example, the Zapier plug-in facilitates integration with a staggering 5,000+ applications, including Google Sheets, Trello, Gmail, HubSpot, and Salesforce. This integration optimizes workflows, supercharges marketing, automates mundane tasks and boosts productivity. As a result, smaller businesses and solopreneurs can allocate their time and resources toward high-impact activities that fuel expansion.

App integrations

Challenges

Early Signals

Potential data misuse

The magnitude of personal information an AI shopping assistant might collect and put to use is staggering. Virtual assistants would be privy to a wealth of sensitive data, from biometric data for apparel selection to medical records for filling prescriptions to financial information. The AI shopping bot could make suppositions about our ideological affiliations by tracking our consumption patterns. Under these circumstances, it’s possible to imagine significant privacy and security risks and potential misuse or exploitation of data. All of this could lead to regulatory scrutiny.

Numerous platforms are integrating with ChatGPT. Klarna has announced a plug-in that lets users solicit shopping advice and get product recommendations via an AI-powered chatbot. Instacart, the grocery delivery service, plans to deploy AI bots to offer recipe ideas and autonomously procure the needed ingredients.

Harnessing synergies

Challenges

Early Signals

Inherent biases

One of the perennial challenges associated with AI is bias, and generative AI is not immune. Often reliant on historical data, generative AI might inadvertently favor destinations already entrenched in the travel pantheon, as voluminous data exists regarding them. Conversely, it might neglect emerging destinations that deserve exploration if the data associated with them is sparse or negatively skewed.

Generative AI is steadily permeating the travel and hospitality sector, with companies like Expedia, Kayak and OpenTable incorporating AI-powered bots that streamline travel planning and automate restaurant reservations. These initial forays into AI point toward a future in which this technology delivers tailored recommendations, intelligent itinerary construction, real-time assistance and frictionless booking.

AI-powered travel bots

Companies are exploring how to employ generative AI to innovate. An initial focus area for many is to seek ways to leverage the technology with human oversight to reduce risks while improving internal systems and operations.

Consumer and business demand for security, personalization, convenience and automation will likely be focus areas for generative AI users. For example, financial institutions will leverage it for personalized banking services, fraud detection and regulatory compliance. Gen AI is instrumental in risk mitigation, as it can generate synthetic data sets to test the efficacy of fraud detection algorithms. Retail and e-commerce companies can employ generative AI to optimize supply chains, bolster marketing efforts through hyper-personalization, improve measurement, and accelerate speed-to-market for new products and services.

Any powerful new technology gives rise to ethical questions and warnings about its misuse. Generative AI is no exception. While some anxieties may be excessive, addressing the risks associated with AI is essential to ensure responsible usage. Critical issues to address include:

The amplification of bias

The spread of fake information

Generative AI's capability to produce cloned human voices and hyper-realistic media is another concern, especially given how rapidly this content can spread. Social media algorithms could reinforce the power of AI-created fakes. This could erode social trust to a new level. Combatting these challenges necessitates advancements in deep fake detection, responsible AI governance, transparency in social media algorithms, and media literacy education.

Surge in cybercrime

Generative AI is a double-edged sword in cybersecurity. Cybercriminals exploit it to automate attacks, craft phishing campaigns, and target victims across borders. It facilitates social engineering-based attacks, mimicking human communication and operating at an unprecedented scale and speed. It can disproportionately affect vulnerable segments of society. On the other hand, defenders use it to detect vulnerabilities and identify threat patterns. Generative AI can play a role in learning from and identifying malicious code and in training models to block security breaches.

Breaching privacy rights and copyright

Generative AI could violate privacy rights as it scrapes information from the internet and re-uses information submitted in prompts to train its models. That information may include personal data subject to the EU’s General Data Protection Regulation and other privacy laws around the globe. Authorities are already assessing the privacy issue, focusing on the need for transparency, consent and control, data minimization, data accuracy and fairness. Getting this issue right is essential to fostering social trust. A related issue is copyright. Much of the online information that generative AI puts to use is subject to copyright protections, potentially putting generative AI solutions providers in breach.

Job disruption

Generative AI could jostle the employment economy, particularly in white-collar sectors. A recent McKinsey report says generative AI could automate tasks consuming 60% to 70% of some employees' time. If that happens, companies may require fewer workers. While AI could eliminate specific jobs, it will likely create new opportunities and free workers from rote tasks so they can do more strategic work. It is therefore critical to adopt a “people-first strategy” and help workers upskill.

Regulators and innovators

Regulators

Generates video based on text prompts. Valuation: $1.5 billion

Produces email, blog and other content. Valuation: $1.5 billion

Its code-writing app is called Ghostwriter. Valuation: $1.2 billion

Its chatbot is a less expensive, lighter-compute model. Valuation: $4.4 billion

A personal assistant that can complete multi-step online tasks. Valuation: $1 billion

Creates AI-powered persona. Valuation: $1 billion

Stable Diffusion generates images based on text prompts. Valuation: $1 billion

Apps for video and image editing. Valuation: $1.8 billion

Open-source app-building tools. Valuation: $2 billion

Leverages enterprise data to unlock powerful AI applications. Valuation: $7.3 billion

Solutions for document summary and text generation. Valuation: $2 billion

With ChatGPT, it made generative AI a household term. Valuation: $29 billion

China

Individual states have taken regulatory measures and, at the federal level, Senate leaders have unveiled a framework for AI regulation that foresees independent expert reviews and user access to results.

U.S.

The EU’s pioneering Artificial Intelligence Act, expected to be finalized in 2023, is meant to balance AI’s benefits with the need to safeguard the public interest. It would ban facial recognition technology in public spaces and mandate greater oversight of generative AI.

Innovators

Besides contributing significantly to AI research in general, Microsoft made an early strategic investment in OpenAI, securing an exclusive partnership.

Known for ambitious projects, Google co-developed transformer technology in generative AI. Recently, the company has invested heavily in LLMs and AI agents.

Founded by ex-OpenAI staff, Anthropic aspires to be an open and ethically conscious AI research organization. They recently released Claude2 first in the US and UK.

Gaining prominence thanks to its Stable Diffusion image-creation engine, Stability seeks to embody the open-source philosophy in AI.

As the AI landscape evolves, several key players have come to the fore, including:

Giant tech players with extensive resources are making hefty investments to gain an early edge in AI and forge partnerships with startups.

Allocating $100 million to a center dedicated to assisting companies with generative AI.

Investing $3 billion in AI over three years.

Committed to investing $500 million in generative AI startups.

Plans to spend $33 billion supporting "ongoing build-out of AI capacity."

Unicorn startups

The Next (AI) Economy

Data differentiators

Bespoke AI

While API-based plug-ins can make generative AI more useful, open standards will help create specialized AI solutions for specific sectors, such as healthcare, legal, finance and architecture.

Widespread integration

AI-to-AI interactions

Bespoke AI might necessitate a single personalized AI bot to orchestrate other bots. Machine-to-machine payments are increasingly common; in the future we’ll likely see AI-to-AI commerce, where an AI assistant interacts with AI services throughout a transaction to tailor the purchase, schedule delivery and coordinate payment.

Staying power

Unlike other technologies that have seen hype cycles, generative AI exhibits clear use cases, has led to the creation of robust solutions, and is developing swiftly. New opportunities will continue to appear. This technology is poised to be transformative across nearly every sector.

General-purpose models will become commonplace as LLMs integrate into applications, freeing workers from repetitive work so they can exercise their creativity. At the same time, LLM products will be customized to better serve specific sectors.

Reimagining Money

A tokenized world

Accelerating collaboration and insight sharing

Automating software and product development

A Deloitte survey revealed that business leaders consider culture and leadership, operations, and tech and talent to be key to using AI effectively. Eighty-two percent of them believe AI enhances job satisfaction and performance.

Strong enterprise motivation

of global business leaders consider AI critical to success

The global AI market reached nearly $60 billion in 2021 and is projected to grow at a compound annual growth rate of 39.4% to reach $422 billion by 2028. While North America currently holds the largest market share at around 43%, the Asia Pacific region is experiencing faster growth. Industry players will see an uptick as generative AI-related opportunities come online.

Scholastic abilities

CHAT GPT

PerCENTILE

GPT-4

PerCENTILE

GPT-4 (with VISION)

The number of patents filed for AI technologies has soared. In 2021 alone, more than 140,000 AI-related patents were filed globally, a 14-fold increase from about 10,000 in 2017.

Increase in AI patents

AI TECHNOLOGY PATENT IN 2021

KNOWLEDGE DISTRIBUTION

CODE WRITING

Operations and efficiency

Customer applications

Building hyper- personalized content at scale

Marketing

Smart conversational agents for B2B and B2C

CUSTOMER INTERFACES

Optimizing resources to improve services

SERVICE DeliveRy

Strengthening finances via resource allocation

Adaptable programs to upskill employees

HR AND TRAINING

TREasury

Expedited contracts through automation

Nextgen fraud detection and prevention tools

CYBER SECURITY

LEGAL

A generative AI solution is a prediction engine: a machine that (in the case of a text-based model) chooses the next word in a sequence based on patterns identified in the oceans of writing on which it's been diligently trained. The results it produces can be, and increasingly are, startling. But it's incapable of the of the intuition of every human being.

How smart?

IN 2022

Global corporate investment in AI has grown 13-fold over the past decade, from $14 billion in 2013 to nearly $190 billion in 2022.

Surge in AI investment

Challenges

Early Signals

Lack of transparency

In a democratic society, decision-making should be made in a clear, open manner, wherever possible. Turning decisions over to AI could complicate that. For example, the reasons an AI solution assigns us a higher insurance rate will be hidden deep inside its neural networks, and difficult to access even for the solution’s creators. Customers of travel websites might never know why chatbots suggest one destination instead of another. The same goes for users of financial platforms’ automated advisors. This “black box model” could degrade social trust.

Super-tech at the supermarket

Brilliant Labs, a Singapore-based startup that develops AR wearables with the use of ChatGPT, has garnered interest among key Silicon Valley investors such as the co-founders of Siri and Oculus. In February 2023 it launched a monocle-like AR device that users can clip onto their eyewear so that they can interface with ChatGPT and other generative AI apps. Since then, it has raised over $3 million in a seed round.

Interface explorations

In non-anglophone markets, the fact that early generative AI solutions have been trained on English-language data can create challenges. Answers provided by ChatGPT in languages other than English may be rendered in ways that do not sound credible to native speakers, degrading confidence in the solution. In South Korea, just 26.5% of respondents considered ChatGPT answers trustworthy. Naver, South Korea's leading internet conglomerate, is thus keen to develop more localized AI applications for its domestic market and other regions, such as the Arab world and Latin America. With such initiatives in motion, and with Google's Bard chatbot adding services in multiple foreign languages, the lost-in-translation problem is expected to be soon resolved.

Lost in translation?

Security and privacy

In the form of best-in-class security and privacy practice.

Increased accessibility combined with valuable interconnected data will likely accelerate the potential of generative AI.

Active investors

with editing support from GPT4 and design support from MidJourney

Businesses are exploring how to weave this fast-evolving technology into their operations.

As new opportunities emerge, it's essential to balance the technology's revolutionary potential against a formidable set of inherent risks and challenges.

Collaboration, transparency and thoughtful regulation will be vital in navigating the evolving landscape of generative AI so the technology promotes human prosperity rather than working against it.

Generative AI promises to transform customer experiences, facilitate personalized interactions and redefine industries.

Regulatory frameworks and guardrails are crucial, just as they are for any potentially transformative technology. So far, regulators and legislators around the world have struggled to keep up with AI’s advances, but the startling advent and swift progress of generative AI seem to have concentrated their attention. Interestingly, companies are advocating for governmental oversight of AI activities. This suggests that governments, the private sector, and civil society can collaborate fruitfully in developing AI regulation.

Transparency and control

AI solution providers must explain how they use and share individuals’ data and give individuals the ability to control its use.

Accountability

Individual interests must stay at the center of generative AI data practices.

Inclusion

Data practices, analytics and outputs must be inclusive, comprehensive and equitable.

Integrity

AI stakeholders must be deliberate in their actions to minimize biases, inaccuracies and unintended consequences.

Innovation

Better generative AI experiences, products and services must emerge continuously.

Social impact

Generative AI must be a force for good, making a positive impact on communities and individuals around the world.

Enterprise-ready: super tools for companies

LLaMA, Meta’s LLM, is much smaller than its rivals, but its open source architecture could drive significant adoption.

U.S.

China

Near-term explorations

BY 2028

Generative AI can contribute powerfully to human progress. But for it to do so, we’ll have to use it responsibly and with a view toward mitigating unintended consequences.

Regional regulatory landscapes:

Sources 1. GPT-4 has more than a trillion parameters 2. OpenAI- GPT-4 3. The ultimate search engine - Google predicts the future 4. The state of AI in 2022—and a half decade in review 5. The Age of AI has begun 6. For B2B generative AI apps, is less more? 7. $422.37+ Billion Global Artificial Intelligence (AI) Market Size Likely to Grow at 39.4% CAGR During 2022-2028 8. COVID-19 Boosted the Adoption of Digital Financial Services 9. Morgan Stanley Wealth Management Announces Key Milestone in Innovation Journey with OpenAI 10. ‘State of AI in the Enterprise’ Fifth Edition Uncovers Four Key Actions to Maximize AI Value 11. ChatGPT could cost over $700,000 per day to operate. Microsoft is reportedly trying to make it cheaper. 12. ChatGPT Plugins Marketers Rave About 13. Carrefour integrates OpenAI technologies and launches a generative AI-powered shopping experience 14. Instacart, Klarna, Shopify deepen ties with ChatGPT 15. Expedia Releases ChatGPT-Powered AI Chatbot on Mobile App 16. Welcome, robots: KAYAK is now integrated on ChatGPT 17. NEW: ChatGPT restaurant recs, powered by OpenTable 18. Fake ChatGPT apps preying on users’ trust in South Korea 19. South Korea’s Naver to target foreign governments with latest ChatGPT-like AI model 20. How many languages does ChatGPT support? The Complete ChatGPT language list. 21. Deepfakes are biggest AI concern, says Microsoft president 22. The economic potential of generative AI: The next productivity frontier 23. Schumer Lays Out Process to Tackle A.I., Without Endorsing Specific Plans 24. The UK's approach to regulating AI 25. EU AI Act: first regulation on artificial intelligence 26. UAE Government - Generatvie AI Guide 27. South Korea to set new standards for copyrights of AI-generated content 28. China finazlizes rules on generative AI amidst a rush of product launches 29. There is only one question that matters with AI 30. The AI Nonprofit Elon Must founded and quite is now for profit 31. Microsoft's $13 billion bet on OpenAO carries huge potential along with plenty of uncertainty 32. Google open sources trillion parameter AI Language Model switch transformer 33. Alphabet-backed Anthropic outlines the moral values behind its AI bot 34. We are the world’s leading open source generative AI company 35. Cause for a LLaMA? Meta reckons its smaller text-emitting AI is better than rivals 36. Our World in Data 37. Corporate Investment in AI is 13 Times Greater than a Decade Ago 38. CB Insights: The complete list of Unicorn companies (May 31 2023) 39. OpenAI closes $300M share sale at $27B-29B valuation 40. AI Startup Anthropic Raising Another $300M At $4.1B Valuation — Report 41. [CEROS OBJECT] 40. Generative A.I. Start-Up Cohere Valued at About $2 Billion in Funding Round 41. Hugging Face reaches $2 billion valuation to build the GitHub of machine learning 42. Lightricks Valued At $1.8 Billion After Latest $130 Million Funding Round 43. AII company Runway valued at $1.5 billion in latest funding - source 44. AI content platform Jasper raises $125M at a $1.5B valuation 45. Replit, the web-based IDE developing a GitHub Copilot competitor, raises $100M 46. Microsoft-backed AI startup Inflection raises $1.3 billion from Nvidia and others 47. AI Startup Adept Raises $350 Million at $1 Billion Valuation 48. Ex-Google employees’ A.I. chatbot startup valued at $1 billion after Andreessen Horowitz funding 49. Stability AI, the startup behind Stable Diffusion, raises $101M 50. CB Insights: The complete list of Unicorn companies (May 31 2023)

Enterprise

Finance

Small Business

Retail

Travel

Banks

Merchants

Media + Entertainment

Rapid industry growth

In June 2023 Carrefour, the world’s eighth-largest retailer, launched new services powered by OpenAI systems, with support from Bain & Company. The new services include the Hopla chatbot, which helps consumers fulfill their shopping needs and reduces food waste by making suggestions on reusing ingredients and offering recipes. Other services include description sheets that offer customers additional info about products and an integrated gen AI tool that improves internal purchase processes.

Made 18 investments in venture capital-backed startups since the start of 2021.

S.Korea

Early on, South Korea called for the evaluation of generative AI’s impact on society and data privacy. In February 2023, it passed the Law on Nurturing the AI Industry and Establishing a Trust Basis. A few months later, the government set new standards for the copyright of AI-generated content to minimize disputes.

S.Korea

The UK has opted to not regulate AI under a single function. Instead, it is taking a sector-specific approach designed to provide more clarity and to encourage innovation. Regulators intend to provide non-statutory guidance including templates and standards.

UAE

In contrast to jurisdictions setting rules and limitations for the use or AI, the UAE seeks to educate the territory about generative AI and highlight certain platforms and case studies. Its AI Guide calls for more Arabic language applications in natural language processing tools.

UAE

What is GPT?

How does it work?

How smart is it?

Is it sentient?

What are parameters?

The intelligence exhibited by Large Language Models (LLMs) should be understood as pattern recognition rather than true understanding. These models recognize patterns from data inputs and generate responses based on those patterns, which can give the illusion of intelligence. However, they are limited to creating variations within the known patterns and cannot generate entirely new content. Ensuring the accuracy of LLM outputs remains a significant challenge. These models and chatbots can provide false information, hallucinations, nonsensical advice, or fabricated details. To improve their reliability, integrating real-time data sources and curating closed datasets can be beneficial. Additionally, developing specialized "vertical" search engines for specific professional domains can enhance the relevance and accuracy of the information provided by LLMs.

How smart is it?

What is GPT? Neural network models capable of creating human-like text and content and engaging in conversational question-answering. How does it work? GPT models are trained on data and develop responses to user queries based on billions of parameters learned during the training process. What are parameters? Parameters are variables within an AI system that determine how data is transformed into the desired output. How smart is it? It uses pattern recognition that utilizes various learning approaches, including unsupervised methods, during its training process. Is it sentient? Despite producing human-like responses, large language models are not sentient and will not evolve into Artificial "General" Intelligence.

click for more

AI-generated fakes could have world-shaking consequences. Imagine an entirely credible-looking fraudulent video depicting a devastating terror attack on a world financial center. The market crashes; the manipulators, having shorted it, garner a huge profit. In May 2023, the market was rattled by a counterfeited photo of an explosion near the Pentagon. Vigilance against such manipulations will be needed.

New frontiers in market manipulation?

Brad Smith President, Microsoft

Brad Smith, the president of Microsoft, has said that his biggest concern around artificial intelligence is deepfakes, content that looks realistic but that’s false.

Roger McNamee Venture Capitalist

“The question we should be asking about artificial intelligence — and every other new technology — is whether private corporations should be allowed to run uncontrolled experiments on the entire population without any guardrails or safety nets.”

Generative AI employs machine learning techniques on large datasets, enabling models to identify patterns, styles, and connections. Transformer models, developed in 2017 by Google and the University of Toronto, serve as the foundation for generative AI. These models utilize an attention mechanism to focus on essential features in input data, recognizing language or image patterns. Generative AI enables users to generate new content based on a variety of inputs quickly. Inputs to these models can handle structured, unstructured, and semi-structured data, adapting to domains like speech recognition, image processing, audio generation and more. Outputs include text, imagery, audio, 3D and synthetic data.

How does it work?

2017

2022

More firms adopting AI

Percentage of companies who reported using AI for at least one business application. These numbers, for all AI types, portend a bright future for generative AI.

Bill Gates

“The development of AI is as fundamental as the creation of the microprocessor, the personal computer, the Internet and the mobile phone. ... Entire industries will reorient around it. Businesses will distinguish themselves by how well they use it.”

Bobak Tavangar, CEO and co-founder of Brilliant Labs

“This age of AI is upon us, and from day one – even before GPT – we saw a market opportunity with computer vision.”

Alexandre Bompard, Chairman and CEO of Carrefour

“By pioneering the use of generative AI, we want to be one step ahead and invent the retail of tomorrow.”

Generative AI is considered “narrow” and should not be conflated with Artificial “General” Intelligence (AGI). AGI is the type of AI that can learn and reason like we humans do. This means it would have the potential to solve complex problems and make decisions independently. A sentient form of AI would be equipped to process, evaluate, and feel separately from the data it has inputted to come to its own conclusions about a given topic. Opponents cite that this technology poses existential threats, as depicted in movies like Terminator. Yet AGI remains far from realization, requiring substantial resources and policy frameworks to ensure responsible development and deployment. While challenges specific to generative AI demand attention, they differ from AGI concerns.

Sentient AI?

ChatGPT 2 was trained on just 1.5 billion parameters. In contrast, ChatGPT-3.5 boasted 175 billion parameters, producing more intelligent responses across different writing styles. While OpenAI has yet to confirm this, some experts have predicted that the latest upgrade, ChatGPT 4, was trained on over a trillion parameters. GPT-4 enables multimodal input, allowing chatbots to receive information through multiple channels, including text and images. Meanwhile, diffusion models progressively create or degrade images and have shown remarkable success in image generation.

What are parameters?

the number of languages ChatGPT can understand and generate text in

The Microsoft-funded breakthrough start-up whose ChatGPT made “generative AI” a household term, OpenAI is both commercial shop and non-profit dedicated to promoting ethical AI.

The Scale platform provides enterprise data labeling and management services so that businesses can build powerful and customized generative AI models.

Solutions for document summary; generation of email, marketing, blog and other text; semantic search (search based on intent and meaning, not just on keyword-matching) and other tasks.

At once an open-source ML-building tools repository and a knowledge-sharing and collaboration hub for the AI community, this “GitHub of machine learning” in spring 2022 launched HuggingChat, an open-source chatbot.

Lightricks has been making photo/video apps, most famously phone-based editing app Facetune, since 2013. In 2022 it began incorporating AI into its suite of products, now including an influencer marketing platform.

Runway’s collection of video and photo creation and effects tools was augmented in spring 2023 with its new model, Gen 2, another in a small group of solutions that generate video on the basis of a mere text prompt.

Geared for businesses, Jasper cranks out email, blog and other content using simple prompts. It integrates an SEO platform to track how content stacks up and a plagiarism screener to ensure it’s on the level.

The company’s Ghostwriter app autocompletes code, generates code in line with suggestions, helps modernize code to meet standards and offers users explanations about bits of code.

The Google-backed company claims its Claude chatbot will be “less likely to produce harmful outputs” than similar solutions. Claude Instant is a less-expensive, lighter-compute model.

Its ACT-1 model can complete multi-step online and software-based tasks for you, like a real-life assistant. Think of searching for vacation spots and then booking a trip or renewing a driver’s license.

Check off the attributes you’re looking for and create an AI persona to interact with. It can be personal assistant, a fictional character brought to existence, a representation of a real-life person and more.

Along with ChatGPT, this outfit’s Stable Diffusion image-creation engine, powered by text prompts, was one of the apps that drove home generative AI’s power to the wider public.

Glean’s enterprise AI solutions are geared toward companies’ accounting and financial planning and analysis teams, helping them with AP/AR and other spend management-related functions.

Force multipliers

click for more

GPT-4

PerCENTILE

GPT-4 (with VISION)

PerCENTILE

Scholastic abilities

GPT-4 outperforms ChatGPT by scoring in higher approximate percentiles among test-takers.

Uniform bar exam

Biology olympiad

10th

CHAT GPT

PerCENTILE

90th

31st

CHAT GPT

99th

parameters

1.5

GPT 2

175

GPT 3.5

parameters

trillion

GPT 4

parameters

ARTIFICIAL INTELLIGENCE

MACHINE LEARNING

DEEP LEARNING

Predictive Analytics

NEURAL NETWORKS

GENERATIVE AI

As opposed to machine learning where algorithms learn and resolve problems based on typically, linear progressions for one, deep learning can structure multiple layers of algorithms in its computations, identifying ways to cluster data and apply various weights. This ability to handle and train under these forms offer greater accuracy as more data is inputted.

Deep learning

A method of machine learning that is purposed to simulate the basic workings of a human brain. Similar to that of a network where nodes represent various data inputs that are intertwined with other nodes, neural networks enables the processing of data in such a way it also accounts for associations and connections. This form of AI helps interpret sensory data by clustering and classifying real-world inputs, be it images, sound and/or text that need to be translated in a systematic manner so that models can then be trained on. Neural networks also form the foundation for which generative AI models are based on.

Neural networks

relies on Large Language Models (LLMs), the prevailing form of AI, to create content based on existing data in response to user prompts. It leverages neural networks to identify the patterns and structures within existing data and uses that to generate new and original content. Generative AI has surged in popularity, with around 170 million search results for "generative AI" as of June 2023, suggesting widespread interest and adoption of this transformative technology.

Generative AI

models focus on historical data to make predictions about future outcomes. Predictive analytics uses Machine Learning and AI as tools to ingest and establish patterns in the data to forecast possible results.

Predictive AI

What is GPT?

How Smart?

As VC and enterprise investment ramps up, a growing number of billion-dollar AI startups are emerging.

music

image

text

video

programming code

voice

FAQs

Gen AI can handle multiple modalities of inputs and outputs

GPT, or Generative Pretrained Transformer, is a type of artificial intelligence model developed for natural language processing tasks. It is designed to generate human-like text by predicting the likelihood of a word given previous words used in the text.

What is GPT?

Generative

Pretrained

Transformer

Sources 1. GPT-4 has more than a trillion parameters 2. OpenAI- GPT-4 3. The ultimate search engine - Google predicts the future 4. The state of AI in 2022—and a half decade in review 5. The Age of AI has begun 6. For B2B generative AI apps, is less more? 7. $422.37+ Billion Global Artificial Intelligence (AI) Market Size Likely to Grow at 39.4% CAGR During 2022-2028 8. COVID-19 Boosted the Adoption of Digital Financial Services 9. Morgan Stanley Wealth Management Announces Key Milestone in Innovation Journey with OpenAI 10. ‘State of AI in the Enterprise’ Fifth Edition Uncovers Four Key Actions to Maximize AI Value 11. ChatGPT could cost over $700,000 per day to operate. Microsoft is reportedly trying to make it cheaper. 12. ChatGPT Plugins Marketers Rave About 13. Carrefour integrates OpenAI technologies and launches a generative AI-powered shopping experience 14. Instacart, Klarna, Shopify deepen ties with ChatGPT 15. Expedia Releases ChatGPT-Powered AI Chatbot on Mobile App 16. Welcome, robots: KAYAK is now integrated on ChatGPT 17. NEW: ChatGPT restaurant recs, powered by OpenTable 18. Fake ChatGPT apps preying on users’ trust in South Korea 19. South Korea’s Naver to target foreign governments with latest ChatGPT-like AI model 20. How many languages does ChatGPT support? The Complete ChatGPT language list. 21. Deepfakes are biggest AI concern, says Microsoft president 22. The economic potential of generative AI: The next productivity frontier 23. Schumer Lays Out Process to Tackle A.I., Without Endorsing Specific Plans 24. The UK's approach to regulating AI 25. EU AI Act: first regulation on artificial intelligence 26. UAE Government - Generatvie AI Guide 27. South Korea to set new standards for copyrights of AI-generated content 28. China finazlizes rules on generative AI amidst a rush of product launches 29. There is only one question that matters with AI 30. The AI Nonprofit Elon Must founded and quite is now for profit 31. Microsoft's $13 billion bet on OpenAO carries huge potential along with plenty of uncertainty 32. Google open sources trillion parameter AI Language Model switch transformer 33. Alphabet-backed Anthropic outlines the moral values behind its AI bot 34. We are the world’s leading open source generative AI company 35. Cause for a LLaMA? Meta reckons its smaller text-emitting AI is better than rivals 36. Our World in Data 37. Corporate Investment in AI is 13 Times Greater than a Decade Ago 38. CB Insights: The complete list of Unicorn companies (May 31 2023) 39. OpenAI closes $300M share sale at $27B-29B valuation 40. AI Startup Anthropic Raising Another $300M At $4.1B Valuation — Report 41. Microsoft-backed AI startup Inflection raises $1.3 billion from Nvidia and others 42. Generative A.I. Start-Up Cohere Valued at About $2 Billion in Funding Round 43. Hugging Face reaches $2 billion valuation to build the GitHub of machine learning 44. Lightricks Valued At $1.8 Billion After Latest $130 Million Funding Round 45. AII company Runway valued at $1.5 billion in latest funding - source 46. AI content platform Jasper raises $125M at a $1.5B valuation 47. Replit, the web-based IDE developing a GitHub Copilot competitor, raises $100M 48. AI Startup Adept Raises $350 Million at $1 Billion Valuation 49. Ex-Google employees’ A.I. chatbot startup valued at $1 billion after Andreessen Horowitz funding 50. Stability AI, the startup behind Stable Diffusion, raises $101M

US - Individual states have taken regulatory measures and, at the federal level, Senate leaders have unveiled a framework for AI regulation that foresees independent expert reviews and user access to results. UK - The UK has opted to not regulate AI under a single function. Instead, it is taking a sector-specific approach designed to provide more clarity and to encourage innovation. Regulators intend to provide non-statutory guidance including templates and standards. EU - The EU’s pioneering Artificial Intelligence Act, expected to be finalized in 2023, is meant to balance AI’s benefits with the need to safeguard the public interest. It would ban facial recognition technology in public spaces and mandate greater oversight of generative AI. UAE - In contrast to jurisdictions setting rules and limitations for the use or AI, the UAE seeks to educate the territory about generative AI and highlight certain platforms and case studies. Its AI Guide calls for more Arabic language applications in natural language processing tools. South Korea - Early on, South Korea called for the evaluation of generative AI’s impact on society and data privacy. In February 2023, it passed the Law on Nurturing the AI Industry and Establishing a Trust Basis. A few months later, the government set new standards for the copyright of AI-generated content to minimize disputes. China - In April 2023, China drafted rules that required companies to register generative AI products and submit them for security assessments. More recently, it has adopted a softer approach designed to encourage Chinese companies to compete globally.

Use cases include stand-alone generative AI integrations and also making traditional AI applications more compelling and personalized with a gen AI overlay.

Transformer This is the type of model architecture used by GPT. Transformers use a mechanism called attention, weighing the importance of different words in a sentence when generating predictions.

Generative This refers to the model's capability to generate new, creative outputs, such as writing a text, given some input.

Pretrained This refers to the model's initial training phase, where it learns from a large corpus of Internet text. Pretraining allows the model to learn grammar, world facts, some reasoning abilities, and even biases in the data it was trained on, all without human supervision.