By Christopher Gannatti, CFA
Let’s face it, we love exciting announcements. Why talk about the small technical improvements of a artificial intelligence (AI) system when you can predict the coming advent of artificial general intelligence (AGI)? However, focusing too much on AGI risks missing out on many incremental improvements in the space along the way. It’s a lot like how focusing only on when cars can literally drive themselves runs the risk of missing out on all the incremental assisted driving features being added to cars all the time.
DeepMind in the foreground… AGAIN
The cover of Alpha Go, DeepMindit is1 system that was able to surpass the performance of professional Go player Lee Sedol, was a game-changer. Now there is AlphaZero, Alpha folding and more. DeepMind has made incredible progress showing how AI can be applied to real problems. AlphaFold, for example, predicts how given proteins will fold and, by knowing precisely the shape of given proteins with precision, unlocks enormous potential in the way we think about all kinds of medical treatments.
The Covid-19 vaccine using mRNA was based largely on targeting the specific ‘spike protein’ shape. The overall problem of protein folding was something humans had been focusing on for over 50 years.2
However, DeepMind recently introduced a new “general purpose” AI model called Gato. Think of it this way – AlphaGo is specifically focused on playing Go and AlphaFold is specifically focused on protein folding – these are not general purpose AI apps. On the other hand, Gato can3:
- Play Atari video games
- Caption Images
- Stack blocks with a real robot arm
In total, Gato can perform 604 tasks. This is very different from more specialized AI applications which are trained with specific data to optimize a task.
So, AGI is now on the horizon?
To be clear, the full AGI is a significant leap from anything done to date. It’s possible that with an increase in scale, the path Gato used could lead to something closer to AGI than anything that has been done to date. Likewise, scaling up alone may not get you anywhere. The AGI may require breakthroughs that are not yet determined.
People love to get excited about AI and its potential. In recent years, the development of GPT-3 by Open AI4 was great, as was the image generator SLAB. Both of these accomplishments were enormous, but neither led to a technology exhibiting human-level understanding, and it is unclear whether the approaches used in either will naturally lead to AGI in the future.
If we can’t say when AGI will come, what can we say?
While massive breakthroughs like AGI can be difficult, if not impossible, to predict with certainty, the focus on AI as a whole has grown incredibly. The recently released Stanford AI Index report is extremely helpful, in that we can see:
- The scale of investment poured into the space. Investment, in a sense, partly measures confidence, in that there must be a reasonable belief that productive activity will result from the efforts being funded.
- The scope of AI activities and how the activities universally display improvement metrics.
The growth of investment in AI
Looking at Figure 1, the progression of investment growth has been staggering. We recognize that this is partly driven by the excitement and potential of AI itself, as well as the general environment. The fact that 2020 and 2021 presented such large figures must be influenced by the fact that capital cost was minimal and the money was chasing exciting stories with potential profits release in the future. Based on what we know today, it would be difficult to predict that the 2022 figure would exceed the 2021 figure, but it could still be an absolute strong sum of money.
It is also interesting to consider the evolution of the components of the investment:
- 2014 was defined by public offerwhich in other years was generally on the lower end of the spectrum compared to the totals.
- The main driver of the steady growth in investment was on the private side, so it seems clear that Figure 1 illustrates the cyclical recovery in private investment, which we recognize will not necessarily continue a linear upward trend. throughout the 2020s.
Figure 1: Global business investment in AI by INVESTMENT ACTIVITY, 2013–21
What activities are funded by the money?
Aggregate investment amounts are one thing, but it’s more concrete to look at specific business areas. Figure 2 is helpful in this regard, as it gives an idea of the change in 2021 compared to 2020.
- In 2021, “Data Management, Processing, Cloud”, “Fintech” and “Medical and Healthcare” led the way, each exceeding $10 billion.
- It should be noted that in the 2020 data (purple), “Medicine and healthcare” was in the lead with approximately $8 billion. This highlights the relative year-over-year increase for “data management, processing, cloud” and “fintech.”
Figure 2: Private investment in AI by focus area, 2020 vs. 2021
Is the AI technically improving?
This is a fascinating question, the answer to which can have almost infinite depth and can be covered in any number of academic papers to come. What can be noted here is the fact that these are two separate efforts:
- Design, program or otherwise create the specific AI implementation.
- Determine the best ways to test if it is actually doing what it is supposed to or if it is improving over time.
I find “semantic segmentation” particularly interesting. It sounds like something only an academic would ever say, but it references the concept of seeing a person riding a bicycle in a photo. You want the AI to know which pixels are the person and which pixels are the bike.
If you think – who cares if a sophisticated AI can discern the person on the bike in such an image? – I grant you that it may not have the highest application value. However, imagine an internal organ shown on a medical imaging device – now consider the importance of distinguishing healthy tissue from a tumor or lesion. Can you see the value this could bring?
The Stanford AI Index report actually breaks down specific tests designed to measure the progress of AI models in areas such as:
- computer vision
- Reinforcement learning
- Hardware training time
Many of these fields approach what one might define as the “human standard”, but it is also important to note that most of them only specialize in the specific task for which they were designed. .
Conclusion: it is still early days for AI
With some megatrends, it’s important to have the humility to recognize that we don’t know for sure what’s going to happen next. Within AI, we can predict certain innovations, whether in vision, autonomous vehicles or drones, but we must recognize that the biggest returns may come from activities that we do not yet track. For those interested in an investment vehicle designed to expose themselves to AI, consider the WisdomTree Artificial Intelligence and Innovation Fund (WTAI).
Stay tuned for Part 2, where we discuss recent results from some companies operating in the space.
1 DeepMind is a subsidiary of Alphabet. As of June 23, 2022, Alphabet weighed 1.46% in the WTAI.
2 Source: Sky, Will Douglass. “DeepMind’s Protein Folding AI Has Solved a Grand 50-Year-Old Biology Challenge,” MIT Technology Review, 11/30/20.
3 Source: Melissa Heikkila, “Hype Around New DeepMind AI Models Missing What’s Really Cool”, MIT Technology Review, 5/23/22.
4 OpenAI holds a 0% stake in the WisdomTree Artificial Intelligence and Innovation Fund (WTAI).
Important risks related to this article
Christopher Gannatti is an employee of WisdomTree UK Limited, a European subsidiary of WisdomTree Asset Management Inc.’s parent company, WisdomTree Investments, Inc.
There are risks associated with investing, including possible loss of capital. The Fund invests in companies primarily involved in the investment theme of artificial intelligence (AI) and innovation. Companies engaged in AI typically face intense competition and potentially rapid product obsolescence. These companies are also heavily dependent on intellectual property rights and may be harmed by the loss or degradation of these rights. Additionally, AI companies typically invest large sums in research and development, and there is no guarantee that the products or services produced by these companies will be successful. Companies that capitalize on innovation and develop technologies to replace older technologies or create new markets may not succeed. The Fund invests in securities included in or representative of its index, regardless of their investment merit and the Fund does not attempt to outperform its index or take defensive positions in falling markets. The composition of the Index is governed by an Index Committee and the Index may not perform as intended. Please read the Fund’s prospectus for specific details regarding the Fund’s risk profile.
Christopher Gannatti, CFA, Global Head of Research
Christopher Gannatti started at WisdomTree as a Research Analyst in December 2010, working directly with Jeremy Schwartz, CFA®, Director of Research. In January 2014, he was promoted to Associate Director of Research where he was responsible for leading various groups of analysts and strategists within WisdomTree’s broader research team. In February 2018, Christopher was promoted to Head of Research for Europe, where he will be based in WisdomTree’s London office and will be responsible for all of WisdomTree’s research efforts in the European market, as well as supporting the UCIT platform on a global scale. Christopher came to WisdomTree from Lord Abbett, where he worked for four and a half years as a regional consultant. He received his MBA in Quantitative Finance, Accounting and Economics from NYU’s Stern School of Business in 2010, and he received his BS in Economics from Colgate University in 2006. Christopher holds a Chartered Financial Analyst designation.
Editor’s note: The summary bullet points for this article were chosen by the Seeking Alpha editors.