Announcement: Part One: A Realistic Assessment of Progress in Artificial Intelligence | newsletter

Let’s be honest – we love sexy ads. Why talk about small technical improvements to a particular AI system when you can predict the upcoming arrival of AKI? However, focusing too much on AKI risks missing out on many incremental improvements in this area, such as focusing on when cars can literally drive themselves, resulting in the loss of all the overlooked additional driver aids that are constantly installed in cars.

DeepMind in the lead

Covering AlphaGo, the DeepMind system that outperformed professional Go player Lee Sedol, was a turning point. Now there’s AlphaZero, AlphaFold, and more. DeepMind has taken incredible strides in demonstrating how AI can be applied to real-world problems. For example, AlphaFold predicted how certain proteins would bend. Knowing the exact shape of some proteins has opened up huge possibilities for our deliberations about some medical treatments.

The Covid-19 vaccine, which is made from mRNA, is largely based on the specific spike protein form. The general problem of protein folding has been something people have been interested in for over 50 years.

However, DeepMind recently introduced a new AI model called Gato. Think of it this way: AlphaGo focuses specifically on the game of Go, AlphaFold focuses on protein folding – these are not general AI applications because they focus specifically on a single task. can gato:

Play Atari video games

explained

chat

Real robotic arm block stack

In total, Gato was able to complete 604 missions. This is quite different from more specific AI applications that are trained on specific data to improve only one thing.

Is AKI now on the horizon?

To put it frankly: the full AKI is far superior to anything that has been achieved so far. It is possible that the path taken by Gato more broadly leads to something closer than anything AKI has achieved. But it’s also possible that a larger dimension can lead to nothing. Acute renal failure may require breakthroughs not yet identified.

People love to get “excited” about AI and its capabilities. In recent years, the development of OpenAI’s GPT-3 and DALL-E image generator has been of great interest. Both were great accomplishments, but neither of them led to technology leading to human-level understanding. It is also not known whether the methods used in both could lead to acute renal failure in the future.

If we can’t say when acute renal failure will come, what can we say?

While it’s difficult, if not impossible, to anticipate major breakthroughs like AKI with certainty, the focus on AI in general has seen an amazing comeback. The recently released Stanford AI Indicator Report is very useful because it sheds light on:

1. The volume of investment flowing in this field. In a sense, investment is a measure of “confidence” because there must be a reasonable belief that productive activity can result from effort.

2. The scale of AI activity and the fact that activity consistently has better metrics.

The growth of artificial intelligence investment

A look at Figure 1 below shows that the trend of investment growth is staggering. We know, of course, that this is partly due to the excitement and capabilities of AI itself, but also to the general environment. The fact that 2020 and 2021 produced such high numbers could be because the capital costs were minimal and the money was chasing interesting potentially profitable stories. Based on current knowledge, it is difficult to predict whether the figure for 2022 will be higher than the figure for 2021.

It is also interesting to look at the development of investment components:

2014 was defined by “Overview”, which has generally been less comprehensive in other years than in total.

The main driver of the steady growth in investment has been from the private sector, so Figure 1 clearly shows the cyclical rise in private investment, which, admittedly, will not necessarily show a straight upward trend in the 2020s.

What activities are financed with money?

Total investment amounts are one thing, it becomes more realistic when you look at individual areas of activity. Figure 2 below is helpful in this regard and gives an idea of ​​the changes in 2021 compared to 2020.

In 2021, “data management, processing, and cloud,” “Fintech,” and “medicine and healthcare” will lead the way with more than $10 billion each.

It is noteworthy that in 2020 data (in purple), the medical and healthcare sector tops the list with about $8 billion. Thus, the relative increase compared to the previous year in “data management, processing and cloud” and “financial technology” is more noticeable.

Figure 2: Private AI Investment by Focus Area, 2020 vs. 2021

Is artificial intelligence improving technically?

This is an interesting question, the answer of which can be infinitely deep and can be the subject of an infinite number of scientific papers. We can see here that these are two distinct efforts:

1. Design, code or otherwise create a specific implementation of Artificial Intelligence.

2. We need to know how best to test whether it is really doing what it is supposed to do, or whether it is getting better over time.

I find “semantic segmentation” particularly interesting. This sounds like something only an academic would say, but it has to do with the notion that in the photo you see a person riding a bike. You want the AI ​​to be able to determine which pixels represent a person and which pixels represent a bike.

If you think it doesn’t matter if a high-end AI can tell the person off the bike in such a picture, I’ll admit that this might not be the highest quality application. However, imagine an image of an internal organ on a medical image – now think about the value of the segmentation of healthy tissue versus a tumor or lesion. Can you imagine what value it could have?

The Stanford AI Index report details specific tests designed to measure the progress of AI models in areas such as these:

computer vision

language

language

Recommendations

Learning Enhancement

Hardware training times

Robotics

Many of these areas approximate what might be called “human standards”, but it is also important to note that most only specialize in one specific task for which they are designed.

Conclusion: It’s still too early for AI

With some big trends, it’s important to have the humility to admit that we don’t know for sure what will happen next. With AI we can predict some of the innovations, whether in machine vision, autonomous vehicles or drones, but we also need to recognize that the biggest gains may come from activities we haven’t yet pursued.

Stay tuned for “Part 2” in which we discuss the latest findings from specific companies in the field.

This material has been prepared by WisdomTree and its affiliates and is not intended to be relied upon as forecast, research or investment advice. In addition, it does not constitute a recommendation, offer, or solicitation to buy or sell any security or the adoption of any investment strategy. Opinions expressed are as of the date of manufacture and are subject to change subject to the terms below. The information and opinions expressed in these materials are derived from proprietary and non-proprietary sources. Therefore, WisdomTree and its affiliates, employees, officers and agents have no liability for its accuracy or reliability, nor for any errors or omissions otherwise made (including liability to anyone for negligence). Use of the information in this material is at the reader’s sole discretion. Increases in value in the past do not allow any conclusions to be drawn about future results.

1: Source: https://www.cdc.gov/coronavirus/2019-ncov/vaccines/different-vaccines/mrna.html#:~:text=The%20Pfizer%2DBioNTech%20and%20Moderna,use%20in%20the %20 United %20 States.

2: Source: Hekila, Melissa. “The hype around DeepMinds’ new AI models is missing out on what’s really cool about them.” MIT Technology Review. May 23, 2022.

Wisdom Tree Visions

Image sources: WisdomTree, WisdomTree

Leave a Comment