Vladislav Shershulsky, head of Microsoft’s promising technologies in Russia, continues the tradition of New Year’s technological forecasts and shares his views on the near future.
II after GPT-3
Trend. Artificial intelligence has finally penetrated so widely into our lives that we have ceased to notice it. But I must admit that he is doing well so far with a rather narrow class of tasks of imitating our cognitive abilities – to recognize an image or emotion, color a picture, make an intuitive decision.
It’s a long way from Artificial General Intelligence (AGI). Or it was far from before the demonstration in 2020 by the developers from Open-air of their natural language processing algorithm GPT-3 (Generative Pre-trained Transformer 3). A GPT-3-based system can be mistaken for a human – she writes well the code and even “moonlights” as an editor at The New York Times.
Since the advent of GPT-3, the rules of the AI game have changed irreversibly. If earlier it was mainly a “game of the mind”, and a small team with limited computing resources could achieve success in it, now it is a “big race” in which success is based on developments in the field of architecture and technology of specialized microprocessors, cloud resources for collecting, storing and processing information, as well as – for the achievement of related disciplines, from biology to philosophy.
Forecast. Commercial operation of GPT-3 was planned to begin at the end of 2020, but, most likely, this will happen in 2021. New services will appear based on it. Will they be universal?
If I had to answer in one sentence the question what is missing in GPT-3, I would say that she does not know what she does not know. She perfectly answers correct questions about existing things and, in general, is capable of inferences.
But she is still not doing very well with incorrect questions – she does not have a “sense of reality” – her whole world is a corpus of texts with a volume of about 600 GB. On mission-critical systems, these errors can be costly – a problem that developers will need to address.
The next qualitative breakthrough (development of the conditional GPT-4) will require, apparently, a hundredfold increase in the size of the model and an even more significant increase in the involved computing resources.
According to an optimistic estimate, this milestone can be reached as early as 2021, but it will require large investments and serious developments, including the use of special processors for machine learning like Graph core. If all goes well, we will be able to give such an AI not only individual orders, but also some powers.
“Big small world”
Trend. At first the world was big – it took several years, or even a whole life, to circumnavigate the globe. Then it became small, that is, affordable – once I had to fly around it in 48 hours. And suddenly the world became both large and small at the same time. The pandemic has made its own adjustments: it is still possible to call Kamchatka or Australia, but it is not always possible to go to a neighboring city and just to visit relatives on another street. The habitat of modern man has narrowed and expanded at the same time, and this cannot but affect the development of infrastructure and life support technologies.
Electronic means of communication have acquired an existential meaning far beyond their original utilitarian framework. Now they are both a business rescue tool and a means of preserving family and friendships.
Forecast. Old houses will be re-equipped, and new ones will be built with the expectation of greater autonomy, efficiency and environmental friendliness. Satellite Internet, smart homes and greenhouses, renewable energy sources and, in the long term, cheap delivery by drones, will once again significantly expand the human habitat.
Already in 2021, we will see interesting high-tech projects for the construction of settlements in previously inaccessible places and, possibly, attempts to systematically renovate small towns, the shortcomings of which will be compensated for by a new generation of high-tech infrastructure.
Collaboration and communication tools will rapidly evolve towards “real-time social networking””. For example, Microsoft Teams introduced Together Mode – now video conference participants can literally “sit at neighboring desks.” There is a huge potential for virtual and augmented reality solutions.
Over the past years, such devices have become noticeably more affordable, but there were not enough popular application scenarios all the time. Now there are a lot of them – from buying real estate and advice on self-repairing household appliances to visiting educational laboratories and concerts in unusual formats.
Concert of the legend of electronic music Jean-Michel Jarre in the virtual Notre Dame Cathedral
The next major sporting events will be organized entirely with telepresence in mind. And esports will hit the Olympic Games program much faster than we thought.