Venture Capitalist Marc Andreessen explains how AI can change the world

The recent breakthrough in the field of artificial intelligence and machine learning has allowed computers to understand the world around them and intelligently respond to events around them. Google has already begun to introduce these technologies in Android, but recent innovations open up prospects for much larger-scale scenarios for the use of AI, affecting many areas of activity - from designing drones to making medical diagnoses.

At least, this is the point of view of Marc Andreessen, a prominent venture capitalist and one of the founders of the investment fund Andreessen Horowitz. And such a person knows what he is talking about. He made his fortune 20 years ago by founding Netscape together with his partners, and relatively recently his current company was noted for investing in successful projects such as Facebook, Twitter, Airbnb, Slack and Lyft. Andrissen is in constant contact with entrepreneurs and investors trying to create new large technology companies.

According to the investor, recent advances show that artificial intelligence has the potential to become the basis for the emergence of a new generation of large and important technology companies. At the same time, Andreissen recognizes that certain industries have shown strong and stubborn resistance to technological change and expresses the need to do even more work so that every corner of the economy can feel the full power of modern software.

In late September, the journalist Vox took a telephone interview from Mark Andriessen, presented below. The original recording has been abbreviated and edited for clarity.
')
Timothy Lee: What do you think, in what area should we expect the emergence of new great technology companies? In the 90s, we followed Google and Amazon, and in the 2000s, Facebook and Uber repeated their success. Obviously, there may already be a startup in the world that I haven’t heard about yet and is about to be a great success. And yet, it’s hard for me to imagine that at least one of the companies founded in the last 6 years can grow to the size of Google, Facebook or Amazon.

Mark Andriessen: Traditionally, this happens with new platforms and architectures. Emerging generations of new technologies. The last major category of technology that has shown itself in this way is smartphones and smartphone applications. Smartphones appeared in 2007, while the main categories of applications took shape in 2010 or 2011. Today we can confidently talk about the existence of a whole group of large companies whose business is built around the idea of ​​smartphones and which, quite likely, will soon begin to play an important role. However, 4 or even 2 years ago, this was not nearly as obvious as it is today.



Therefore, if we consider the introduction of smartphone architecture as the last of significant events, now there is a feeling that artificial intelligence, virtual reality, autonomous systems, voice technologies and the Internet of things are candidates for the next wave of innovations. The obvious example now is AI. Looking at it, we can confidently say that we are waiting for the emergence of a new group of products and companies whose work will be based primarily on the use of AI.

Facebook, Google and Amazon have launched a huge first-class activity in this area. But besides them, there is a whole legion of startups. I think this will give rise to a whole generation of new, very important AI companies, many of which are just starting their activity now.

Timothy Lee: We have been hearing conversations about AI for quite some time, and yet so far commercial success has rather turned away from similar developments. Why do you think that now everything will be different?

Mark Andrisn : At first, I was frankly skeptical. Few people remember, but in the 80s there was already one AI bubble. He pouted thanks to a whole bunch of companies that received venture financing, and ultimately they all fell short of expectations and simply burned all the capital.

Now we have the feeling that now we see something different. The really big change was the 2012 ImageNet competition . In that year, computers became better people cope with the recognition of objects in the images. It was a real competition, with clear rules for measuring results.

In fact, over the past four years, we have seen one breakthrough after another. First, a revolution occurred in the field of object recognition in static images. Similar breakthroughs are taking place right now in the work with the recognition of objects in video, and they already lead to the emergence of completely new ways of classifying video. And if you can make video recognition, you can cope with changing the picture in real time, which, in turn, opens up the possibility for you to create autonomous systems.

Among the companies that received our financing is Skydio , which is developing fully autonomous consumer drones. Their product is so different from what is on the market today and opens up new opportunities so much that it’s just something on the verge of the supernatural. He follows you without any human control. You run into the forest, and he makes his way through the branches of trees and flies between them completely independently. Their products will set a new price point for consumers. They seem to have flown to us straight from some science fiction film.

And we are witnessing how in-depth training is used for early diagnosis of cardiac pathologies. Another company in which we invested - Freenome - is engaged in the use of in-depth training for analyzing the results of blood biopsy for the purpose of diagnosing cancer and, so far everything shows that their work gives good results.

Timothy Lee: A classic question for the technology industry: “Is this a new product or is it just a feature?” That is, Google, Facebook, and Amazon are investing a lot of money in artificial intelligence. Siri, for example, started as a startup, but was quickly acquired by Apple. Therefore, will the AI-technology be able to give life to large and independent companies able to introduce new products on the market? Or is it more likely that these innovations will be absorbed by existing large companies in order to improve existing products?

Marc Andreessen : Two years ago I thought that large companies would dominate the market. They then had several major advantages:

• The number of people who know how to create this or that thing is always limited. Large companies can pay them more money than startups. Big companies pay them like sports stars. Therefore, large companies will be able to hire them all, and then there will be no free valuable personnel.
• Such projects are very large and complex. This is a very advanced new technology field. For example, 1500 engineers worked on the Echo project in Amazon for 4 years. None of the start-ups can handle this amount of work.
• An additional snag was the need for data. Then everyone thought you really needed huge datasets. Partly, the breakthrough on ImageNet was due just to the enormous size of the database , based on which participants could train their algorithms. Everyone believed that big companies like Google and Facebook would have access to an endless ocean of information, and start-ups could never compare with them in this regard.

Over the past two years, each of these factors has changed to some degree. Suddenly, you have many more graduates with a degree in computational science who know how to do this or that thing, because it has become a new hot area of ​​computer technology. And you also had many engineers who were engaged in solving these same problems, working on the main players in the industry, and then you realized that they could create their own companies.

Now Google has spun off a whole new generation of startups in the development of autonomous vehicles. Otto (recently acquired by Uber) stood out in particular from the crowd, but there are about six other companies besides it that are working in this direction.

Meanwhile, technology itself is becoming more flexible and accessible to entrepreneurs. Many new and interesting projects that have appeared in recent years no longer need a staff of 1500 people. Now they need only five. Google opened the source of an interesting thing called TensorFlow , which can be compared with building blocks for deep learning. And now we are seeing how various startups pick up these blocks and start applying them wherever possible. Two years ago, they had no such opportunities.

Science itself continues to move forward. People learn to teach deep algorithms on small data sets. We see startups that either find a tricky way to access large data sets, or come up with ways to organize the work of algorithms in such a way that they need only small sets.

Timothy Lee: Many of the useful and practical AI applications look very promising, but they do not promise at all that the business created on the basis of them will grow to large sizes. If we talk about truly great opportunities, such as autonomous cars, then there is the impression that large companies have an advantage that it is unlikely to compete with.

Mark Andriessen: It seems to me that you are still considering this situation in the context of adding AI to an existing product. But this is not really what happens. Now we see completely new types of products appear, the creation of which was previously impossible.

Let's go back to the drone again. You buy a drone today, start managing it yourself and 20 minutes later, break it against a tree. You say to yourself "damn it was cool," and now you have to buy a new one.

The main players in this market have been talking for a long time about adding features called follow me. Drone manufacturers are thinking about its implementation, and projects on Kickstarter will surely promise to add it. However, none of the dozens or hundreds of these companies have so far managed to make it work.

The reason is that this is not a feature, but a completely new architecture. Such drones need to be created on the basis of AI from scratch. DJI and other manufacturers of drones are betting that this is a feature. We put the need to develop a new architecture.

This is an example of a fundamental rethinking of the invention. If our thesis is correct, it turns out that all existing drones will soon become obsolete. They are no longer needed by anyone simply because they are unable to do the really important things.

If you talk with car manufacturers, you will make sure that they all think that autonomy is a feature and that they are going to add it to existing cars. Silicon Valley companies believe that they need a completely new architecture. They think that all the principal provisions in this area need to be turned upside down in order to invent something new.

Timothy Lee: It sounds like a lot of innovation. At the same time, interest rates are low, and the economy as a whole is growing very slowly. In theory, all this means that at a low interest rate, it becomes easier to borrow money or raise capital, which means that we should already see a surge in investment activity. Nevertheless, statistics show that much more money remains in an intact state than is invested anywhere. What do you think can explain this?

Marc Andriessen: Right now there are two kinds of industries. The first are industries that are rapidly adopting technologies and improving their productivity. Televisions, computer equipment, media and food. On Bloomberg, by the way, there was an article about the fact that food prices are plummeting as food production processes become much more sophisticated.

So, you have sectors of the economy in which there is a sharp increase in productivity. Prices fall quickly. People in these industries are concerned that their jobs are about to disappear or go to China, Japan or Mexico. Everyone says that the revolutionary changes in them rolls over, that they are changing too much under the pressure of technology. Silicon Valley children literally plunge the economy into chaos.

And then there are sectors in which prices are skyrocketing: health care, education, construction services, prescription drugs, care for the elderly, and pre-school childcare facilities. There are very few technological innovations here. All of the above are sectors with insufficient growth of productivity, new technologies and fundamental changes. There are monopolies, oligopolies, cartels, government-controlled markets, and price collusion - in short, a complete set of systemic violations leading to a sharp rise in prices.

The government is pouring more and more subsidies into these markets, but due to the low adaptability of such markets, additional funding provokes even higher prices. Just such a situation is happening in the field of higher education.

And so people employed in these sectors resent the lack of productivity growth. Technological growth in them is not fast enough, and we pay too much for their services.

And now try to draw conclusions from all of this and you will get some confusion and a feeling of some kind of sluggish movement. However, this attitude only prevents us from seeing what is happening in its true light.

You have one group of sectors in which prices are falling rapidly, another group where they are growing rapidly. Over time, in such a situation, rising sectors will “eat up” the entire economy. Consumers are now seeing how their income is “eaten up” by health and education.

For me, the problem is obvious: it is the lack of implementation of technology, innovation and fundamental changes in those sectors where prices are growing rapidly. My thesis on this score is that this is not a technological bubble at all, it is a technological failure. Our problem is not that the technology has become too much, or that people have become too passionate about them. The problem is that we didn’t even come close to the level of technology we need. All these cartel-like branches inherited from the past are designed in such a way that it is simply very difficult to make any fundamental changes to them.

Timothy Lee: One of the characteristic features of all such slow-growing industries is the widespread use of manual or mental labor. Most of all you pay for what the other person - a nurse, teacher, nanny - spends with you his time. You are probably familiar with the concept of the disease cost of Baumol , which says that as manufactured goods become cheaper, people begin to spend more of their resources on scarce goods, that is, in this case, on human labor.

Hence the question arises whether this problem is solvable as such. That is, there will always be some labor-intensive industries with slow growth of productivity in the economy, while the costs in them will grow much faster in comparison with other industries.

Mark Andriessen: At the macro level, I agree with this statement. I believe that it very precisely explains what is happening, and fully agrees that the cost disease of Baumol plays an important role in the processes of changing the value of goods and labor. I would like to emphasize that labor-intensive industries, of course, can remain so further simply because everyone thinks that this is the way it should be, but this does not mean that they will move forward at the same time. If you refer to the productivity literature written in the 80s, you will see how firmly the experts of those years were confident that production automation is possible, but retail automation is not. The widespread demand for manual labor in retail was perceived as something proper and unchanging. Distribution of goods was also unthinkable without him. That is, everyone thought that the person filling the shelves with goods, the cashier making out the purchases, or the employee who helps bring the car to the car will not go anywhere and will always accompany the retail trade.



Then computer checkout or laser scanning was considered a serious step forward. However, experience has shown that laser scanning did not help to increase overall performance. Laser scanning took time. In half of the cases it did not work and then you had to check the price manually. Perhaps the performance even fell, because with the laser you did not need to glue the price tag on the product, since you thought that you would not need it.

Therefore, at that time there were many disappointments associated with attempts to increase the productivity of retail. Of course, over the past 20 years, its performance has grown tremendously. First there was Walmart with its modern approach to organizing supplies, and then Amazon appeared. As for the next step, I believe that the transition from physical to software products is the third stage of increasing efficiency. The delivery of audio in MP3 format or streaming is a much more productive way of distributing music than recording it on a CD and then selling it through stores.

As a result, we now have a giant retail industry that once seemed doomed to manual labor alone, but now, however, almost completely automated. And, of course, in the history of its development were, and there will be moments of mass disorder about the disappearance of jobs.

Timothy Lee: But is it really? Today, 5 million Americans are employed in retail stores, and according to forecasts of the Ministry of Labor, this figure will grow by 7% in the next ten years.

Marc Andreessen: That's right. Here all the Luddites are wrong time after time. That is, this is exactly the phenomenon that you just talked about - the increasing value of scarce services. The number of clerks in retail is growing.

Another example of growth is bank tellers , whose number has increased for decades in a row. Soon this figure may still begin to fall. Be that as it may, it is noteworthy that the number of vacancies of bank tellers continued to grow over the past 30 years despite the widespread proliferation of ATMs and online banking. And it happened just for the very reason you were talking about. That is, because quite unexpectedly, banks had the opportunity to distinguish between automated and personalized services and real people became in demand to provide a higher level of service.

Just recently, Vinod Khosla wrote about the fact that the profession of a doctor will be a thing of the past. He believes that computers will significantly surpass people in the field of diagnostics, and doctors will simply have nothing to do. I think he is absolutely wrong. In my opinion, the work of the doctor is changing, gradually turning into an increasingly high-level, highly paid and important profession as the activities of doctors will be complemented by more intelligent computer technologies.

That is why I am so optimistic about the economy. That is why I think that Luddites and those who advocate slow growth are mistaken. In fact, we are able to create unbelievable job opportunities and make huge improvements in productivity. Contrary to popular belief today, these two phenomena do not contradict each other.