- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
OUTFRONT SERIES
+
PRESENT
GENERATIVE AI:
FINDING THE COMPETITIVE EDGE AMID CREATIVE DESTRUCTION
$15.7 TRILLION
THE AMOUNT OF VALUE THAT GENERATIVE AI COULD ADD TO THE GLOBAL ECONOMY BY 2030
Source: PwC, July 2023
The rapid explosion of generative AI for general use has captured the global imagination with what the technology might enable.
It’s expected to have a profound impact on economic growth by exponentially increasing productivity. That’s because everything today – from culture to consumer products – is a product of intelligence. If the excitement around generative AI proves justified, it may unleash the biggest wave yet of what Joseph Schumpeter, the famed Austrian economist, termed “creative destruction.” AI innovation will refresh the economy by destroying traditional businesses, replacing them with something more productive. For investors, though, the key question is: If generative AI chatbots quickly commoditize intelligence, where will competitive advantage lie in a world of endemic generative AI?
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
1
2023’S AI RALLY ANTICIPATES A RAPIDLY APPROACHING FUTURE
When ChatGPT-3, the first generative AI chatbot, was launched by Microsoft-backed OpenAI in November 2022, it quickly became the fastest-growing app ever, attracting an estimated 100 million users within two months.
*Meta's Threads microblogging app has since surpassed it by reaching 100 million users in five days.
Accelerated Adoption - Number of Days to Reach 100 Million Users chart
ChatGPT
61
854
1608
1903
Netflix
7337
Days xxto 1M
Days to 10M
Days to 100M
Source: Reuters, July 2023
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
OpenAI expects this growth to drive rapid financial gains; in a 2022 pitch to its private investors, it estimated reaching $200 million of revenue in 2023, rising to $1 billion in 2024.
With rapid adoption across sectors and use cases, the technology is improving exponentially faster at a rate not previously seen. Indeed, a new wave of AI helpers, or agents, like the Microsoft 365 Copilot are being developed to perform complex personal and work tasks without needing close supervision.
Source: Reuters, December 2022
Rapid Growth of GPT Models
Parameters are one of the defining characteristics of AI models. They determine how the model processes the data and how it generates predictions.
GPT-1 2018
GPT-2 Feb 2019
GPT-3 June 2020
Google Bard March 2023
GPT-4 March 2023
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
120 million
1.5 billion
175 billion
540 billion
More than 1 trillion (estimated)
Source: Open AI [1, Feb. 2019] [2, Nov. 2019] [3, Sept. 2020], Semafor, March 2023, Google, July 2023
But rather than just incorporating more and more data, can generative AI autonomously outperform humans at the most economically valuable work, as outlined in the OpenAI charter? And when? Researchers have put GPT to the test, literally, to measure performance against prospective lawyers.
How GPT Models Performed on the Multiple-Choice Multistate Bar Examination (MBE)
Average MBE Passing Range
Q1 2019
Q4 2022
Q1 2023
Random Guessing
80%
70%
60%
50%
Correct Rate
40%
30%
20%
10%
0%
GPT-2
ada 001
babbage 001
curie 001
davinci 001
GPT-3.5
ChatGPT
GPT-4
Student Avg. (NCBE BarNow)
Source: “GPT-4 Passes the Bar Exam,” Katz, Bommarito, Gao, Arredondo. March 2023
GPT <=3.5
GPT-4
NCBE
Generative AI quickly became 2023’s stock market sensation, sparking a huge rally mainly in the U.S. for shares in Microsoft and other tech companies poised to benefit.
Despite a bleak backdrop of rising interest rates and possible recession, the price of big U.S. tech companies leapt in the first six months of 2023.
U.S. Tech Stock Rally
NVDA
MSFT
META
AAPL
AMZN
GOOG
ORCL
$500
01/03/23
02/06/23
03/11/23
04/13/23
05/17/23
06/19/23
07/23/23
08/25/23
$400
$300
Close Price ($)
$200
$100
0
For illustration purpose only. All investments involve risk, including the possible loss of capital.
Source: Refinitiv
For illustration purpose only. All investments involve risk, including the possible loss of capital.
For illustration purpose only. All investments involve risk, including the possible loss of capital.
So, too, has the price of the Netherlands’ ASML, a leading supplier of the equipment needed for making the semiconductor chips that generative AI requires, as well as the price of Taiwan Semiconductor.
ASML Holding N.V.
Taiwan Semiconductor Manufacturing Co. Ltd. ADR
U.S. :NYSE
U.S. :NYSE
$800
$120
$700
$100
$600
$80
$500
$60
Jan ‘23
Jan ‘23
Feb ‘23
Mar ‘23
Apr ‘23
May ‘23
Jun ‘23
Jul ‘23
Feb ‘23
Mar ‘23
Apr ‘23
May ‘23
Jun ‘23
Jul ‘23
Aug ‘23
Aug ‘23
Source: Refinitiv
THE RALLY’S FLIPSIDE
Beyond the big tech companies, there are already signs of companies being disrupted. For instance, the share price of Chegg, the U.S. edtech company, has fallen by two thirds in the first six months of 2023.
It fell by 50% in one day during May after reporting a hit to its business from generative AI, as some students turned to chatbots for answers rather than subscribing to its services.
Chegg’s AI Reckoning
U.S. : NYSE
$30
$24
$18
$12
$6
0
01/03/23
02/06/23
03/11/23
04/13/23
05/17/23
06/19/23
07/23/23
08/25/23
For illustration purpose only. All investments involve risk, including the possible loss of capital.
Source: Refinitiv
The sectors most at risk of having work tasks automated could also be those most at risk of losing their competitive advantages.
That puts asset-light intelligence-based businesses like law firms, architecture and engineering at the top of the list, according to Goldman Sachs’ insights into where work is most likely to be automated. Similar businesses to education technology – such as translation and financial advice companies – might also prove to be at risk.
Share of Industry Employment Exposed to Automation by AI in the U.S.
50%
40%
30%
20%
10%
0%
46
44
37
36
35
33
32
31
29
28
28
28
27
26
26
19
12
11
9
6
4
1
All industries
Production
Management
Sales & related
Office & admin support
Protective service
Healthcare support
Food prep & serving
Life, physical, social sci
Personal care & service
Legal
Business & financial ops
Edu instruction & library
Farming, fishing, forestry
Computer & mathematical
Construction & extraction
Installation, maint, repair
Art, dsgn, ent, sport, media
Architecture & engineering
Community & social service
Bdg & grounds clean/maint
HC practitioners & technical
Transport & material moving
Source: Goldman Sachs GIR, April 2023
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
2
SUSTAINABLE COMPETITIVE ADVANTAGE IN AN AI WORLD
In a world where all businesses have access to generative AI, how do you sustain competitive advantage? The technology looks likely to disrupt, or at least dilute, many existing economic moats – the characteristics of a business that protect it from competition and maintain its profit margins. AI could reshape entire industries and create completely new ones.
Some investors argue that the big tech companies’ generative AI models are becoming increasingly commoditized and may be outcompeted by smaller open-source models – and the first quasi-autonomous agents may be developed by nimble start-ups. After all, the large language models used by chatbots like ChatGPT are available to everyone.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
In a generative AI world, two new sources of competitive advantage are emerging – the quality of your data and the quality of your people, especially those with advanced AI skills.
Turning first to data, long-term advantages can come from how a business fine-tunes its large language models.
One of the reasons that data is very critical to artificial intelligence, and machine learning in general, is that the models themselves are useless without data. There are sometimes trillions of parameters that need to be estimated, and unless you have enough data to train the models, the output is not going to be useful at all. So data is really critical and people and firms that have large amounts of proprietary data are going to be at a clear advantage in this space. ”
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
Fine-tuning data for an existing open-source or paid model is inexpensive.
$1,915 - $7,418
Cost to fine-tune a large language model to complete a complex legal classification according to a model by Snorkel AI that equips organizations to build AI models. Such an application could save hours of a lawyer’s time, which can cost up to $500 per hour.
Source: Boston Consulting Group, March 2023
Sometimes the data doesn’t need to be proprietary — just very hard to gather.
If, for example, you wanted to use an AI program to automatically identify building materials from photographic images, you would need to take photos of hundreds of thousands of structures and map those structures to the relevant building materials, according to Deloitte. Most of the work in building the AI program would go into gathering and organizing the data. It’s not proprietary but the scale of the effort involved acts as a barrier to entry.
Source: Deloitte, 2023
Turning to people, employees with AI skills who can help a business fine-tune data or integrate AI into its processes can also create a competitive advantage.
The U.S. saw nearly 800,000 AI-related job openings in 2022, according to the Stanford Institute for Human-Centered Artificial Intelligence’s 2023 AI Index Report. And in March 2023, LinkedIn reported that there were just over 100,000 jobs on its site that mentioned artificial intelligence in the job description.
AI Job Postings (% of All Job Postings) by Geographic Area, 2014-22
2.05%, United States
2.00%
1.45%, Canada
1.50%
1.33%, Spain
1.23%, Australia
1.20%, Sweden
1.16%, Switzerland
AI Job Postings (% of All Job Postings)
1.14%, United Kingdom
1.00%
1.01%, Netherlands
0.98%, Germany
0.89%, Austria
0.86%, Belgium
0.84%, France
0.72%, Italy
0.50%
0.45%, New Zealand
0
2014
2015
2019
2020
2021
2022
2016
2017
2018
Source: Lightcast, 2022 | Chart: 2023 AI Index Report
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
Of course, AI will not breach all traditional economic moats.
For example, a company that exists in a business where the start-up costs are prohibitive for new entrants would have a formidable competitive advantage. Similarly, a pharmaceutical company with a patent for a miracle drug would still benefit from the exclusive right to market the drug.
The big companies who are already very, very powerful, who also have access to a lot of data, it's likely that these big tech firms will only grow bigger, more powerful with fewer employees. They can do the same work with fewer employees because they use the same technology…. So they will become leaner and smaller, but also a lot more powerful. And then on the other end, we have the open-source approach. That might also be a company coming from left field. So it's very difficult to see where this will go.”
MARK VAN RIJMENAM
Strategic futurist
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
3
WHICH COUNTRIES ARE GAINING COMPETITIVE ADVANTAGE?
If investment in AI leads to competitive advantage, then the U.S. is far ahead of even China, its nearest rival in the race to become leader.
In 2022, the $47.4 billion invested by the U.S. was roughly 3.5 times the amount invested by the next highest country, China, and 11 times the amount invested by the U.K. Notably, all of the big generative AI releases have come from the U.S.
Private Investment in AI by Geographic Area, 2022
0
5
10
15
25
30
35
40
45
50
20
47.36
13.41
4.37
3.24
3.24
3.10
2.35
1.83
1.77
1.52
1.35
1.13
1.04
0.72
0.61
Source: NetBase Quid, 2022 | Chart: 2023 AI Index Report
United States
China
United Kingdom
Israel
India
South Korea
Germany
Canada
France
Argentina
Australia
Singapore
Switzerland
Japan
Finland
Total Investment (in Billions of U.S. Dollars)
There's no question that the initial breakthroughs and the current leadership as of 2023 in generative AI technology is really with the U.S. Why? In part, when you look at some of the other biggest players in the world that are trying to compete, for instance, for some of the Chinese launch language models, there is far less Internet data in Chinese language. So unless they want to build their launch language models in English, they're starting off with a linguistical disadvantage and a volume disadvantage in the amount of data available to train their models. “And then there is more likelihood of intrusion and heavy-handed regulation in a country like China than there would be in the U.S. But of course, it's contingent on regulation being able to strike a good balance between protecting the public but allowing the development of this very powerful new technology.”
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
However, there are areas where China pulls ahead. For instance, China dominates industrial robot installations, enabled by AI to collaborate with humans in factories. In 2021, China installed more industrial robots than the rest of the world combined.
What’s more, its citizens are among those who feel most positively about AI products and services. A 2022 IPSOS survey found that 78% of Chinese respondents agreed with the statement that products and services using AI have more benefits than drawbacks. By contrast, only 35% of Americans agreed that products and services using AI have more benefits than drawbacks.
Source: Ipsos, January 2022
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
Positive Sentiment Toward Products and Services Using AI by Country
Germany
37%
U.K.
38%
Japan
42%
China
78%
U.S.
35%
Italy
50%
Spain
53%
Saudi Arabia
76%
India
71%
Australia
37%
Source: Ipsos, January 2022
Developed countries are likely to adopt generative AI fastest because wages are highest and so they have most incentive, according to McKinsey. But this will only feed into productivity gains if displaced workers find new activities with at least the same economic output.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
300 MILLION
Estimated number of jobs lost to AI over a 10-year period
Source: Goldman Sachs, April 2023
The potential for AI to replace humans in the workforce inspired a UK-based thinktank called Onward to advocate for a shift in the burden of taxation from labor to capital.
Source: ONWARD, May 2023
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
4
POSSIBLE DRAGS TO DISRUPTION
Of course, the huge excitement around generative AI’s potential is as yet untested. There are many reasons why the future of generative AI may not prove as revolutionary as supposed.
Most important is the technology’s limitations. Ask ChatGPT what its IQ is and it answers the following:
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
Turning to specific problems, it’s fair to say that generative AI has an intellectual property problem.
There’s a flurry of lawsuits accusing generative AI companies of breaching copyright. Among them, three artists have sued Stability AI, Midjourney and DeviantArt in San Francisco for alleged unauthorized copying of their work to train their systems.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
What’s more, amid existential fears about machine super-intelligence and job losses, there is regulation looming in the U.S., EU and China.
A record 37 AI-related laws were passed in 2022, according to Stanford University. The EU’s AI Act aims to be the first comprehensive regulatory scheme.
Source: Reuters, April 2023
CONCLUSION
Looking to the future, companies with a combination of high-quality proprietary data, and with people with advanced AI skills who can fine-tune data or integrate it into business processes, are likely to gain a competitive advantage. By contrast, the asset-light business sectors where there are predictions of job losses such as law firms, architecture or engineering may prove most at risk.
- Chapters
- descriptions off, selected
- default, selected
This is a modal window.
Beginning of dialog window. Escape will cancel and close the window.
End of dialog window.
Yet there are possible hurdles that could slow the growth of AI. Firstly, regulation is even now being crafted but its eventual shape is unclear. And, secondly, there are battles underway over the ownership of intellectual property scraped by AI companies to train their models. 2023’s rapid share price rallies signal the excitement about generative AI. It’s a fast-moving business revolution that looks sure to breach some economic moats while creating an opening for building others.