DeepMind’s rising losses present why it’s troublesome to run an AI analysis laboratory

DeepMind's increasing losses show why it is difficult to run an AI research laboratory

Last week, shortly after DeepMind’s breakthrough in using artificial intelligence to predict protein folding, news came that the UK-based AI company is costing its parent company Alphabet Inc in losses of hundreds of millions of dollars every year.

A technology company that is losing money is nothing new. In the tech industry, there are numerous examples of companies that burned investor money long before they became profitable. However, DeepMind is not a normal company looking to secure a market share. It’s an AI research lab that had to transform itself into a semi-commercial outfit to ensure its survival.

While the owner, who is also the parent company of Google, is currently happy with the cost of DeepMind’s expensive AI research, there can be no guarantee it will last forever.

DeepMind Profits and Losses

DeepMind’s AlphaFold project used artificial intelligence to advance the intricate challenge of protein folding.

After his Annual report DeepMind has been filed with the UK Companies House Register and has more than doubled its sales. It achieved £ 266million in 2019, up from £ 103million in 2018. But the company’s expenses also continue to rise, rising from £ 568million in 2018 to 2018. The company’s total losses increased from £ 470million in In 2018 to £ 477m in 2019.

At first glance, that’s not bad news. Compared to previous years, DeepMind’s revenue growth is accelerating as losses plateau.

Deepmind income and lossesDeepMind’s revenue and losses from 2016 to 2019

However, the report contains a few more important facts. The document mentions “sales and development compensation from other group companies”. This means that DeepMind’s main customer is the owner. Alphabet pays DeepMind to apply its AI research and talent to Google’s services and infrastructure. In the past, Google has used DeepMind’s services for tasks like managing the power grid of its data centers and improving the AI ​​of its voice assistant.

[Read: Meet the 4 scale-ups using data to save the planet]

This also means that there is still no market for DeepMinds AI. If so, it will only be available through Google.

The document also mentions that the increase in costs “is mainly due to an increase in technical infrastructure, personnel costs and other related costs”.

This is an important point. The “technical infrastructure” of DeepMind runs mainly on the huge cloud services from Google and its special AI processors, the Tensor Processing Unit (TPU). The main research area of ​​DeepMind is in-depth learning that requires access to very expensive computing resources. Some of the company’s projects in 2019 included working on an AI system that was played Starcraft 2 and another who played Quake 3, both of which cost millions of dollars in training.

A spokesman for DeepMind told the media that the costs mentioned in the document are also working on AlphaFold, the company’s acclaimed protein folding AI, is another very expensive project.

There are no public details on how much Google DeepMind charges to access its cloud AI services, but it will most likely rent its TPUs at a discount. This means that without Google’s support and assistance, the company’s costs would have been much higher.

Personnel costs are another important issue. While machine learning courses attendance has increased over the past few years, scientists who can study the latest AI research involving DeepMind are very rare. And according to some reports, the top AI talent command seven-figure salaries.

The growing interest in deep learning and its applicability to commercial environments has sparked an arms race between technology companies to attract top AI talent. Most of the industry’s leading AI scientists and pioneers work either full-time or half-time at large companies like Google, Facebook, Amazon, and Microsoft. The tough competition to attract top AI talent had two consequences. First, like any other area where supply does not match demand, it has resulted in a sharp rise in AI scientist salaries. And second, it has drove many AI scientists from academic institutions that can’t afford great salaries to wealthy tech companies that can. Some scientists remain in academia to continue scientific research, but they are too rare.

And without the support of a big tech company like Google, research laboratories like DeepMind can’t afford to hire new researchers for their projects.

While DeepMind is showing signs that its losses are slowly reversing, its growth makes it even more dependent on Google’s financial resources and large cloud infrastructure.

Google is still happy with DeepMind

DeepMind AlphaStarDeepMind has developed an AI system called AlphaStar that can beat the best players in the real-time strategy game StarCraft 2

According to DeepMind’s annual report, Google Ireland Holdings Unlimited, one of Alphabet’s investment branches, waived repayment of intercompany loans and any accrued interest of £ 1.1 billion.

DeepMind has also received written assurances from Google that it will continue to provide adequate financial support to the AI ​​company “for at least twelve months”.

At the moment, Google appears to be happy with the progress made by DeepMind, which is also reflected in the comments made by Google and Alphabet executives.

Sundar Pichai, CEO of Alphabet, said on the quarterly earnings call in July with investors and analysts, “I am very pleased with the pace at which our research and development in AI is advancing. For me it is important that we as a company are up to date and are leaders. I love the pace at which our engineering and R&D teams are working at both Google and DeepMind. “

However, the corporate world and scientific research move at different speeds.

Scientific research is measured in decades. Much of the AI ​​technology used in commercial applications today has been in development since the 1970s and 1980s. Likewise, much of the latest research and techniques showcased at AI conferences today are unlikely to hit the mass market in the years to come. The ultimate goal of DeepMind is development artificial general intelligence (AGI) is at least decades away according to the most optimistic estimates.

On the other hand, the patience of shareholders and investors is measured in months and years. Investors detest companies that can’t make a profit in years’ time, or at least show hopeful signs of growth. DeepMind doesn’t currently have any of these. It has no measurable growth as its only customer is Google itself. And it’s not clear when – if at all – some of its technology will be ready for commercialization.

Sundar PichaiSundar Pichai, CEO of Google, is pleased with the pace of AI research and development at DeepMind

And here lies DeepMind’s dilemma. At its core, it is a research laboratory that aims to push the boundaries, push the science and make sure advances in AI are beneficial for all people. However, the owner’s goal is to develop products that will solve certain problems and make a profit. The two goals are diametrically opposed and pull DeepMind in two different directions: maintain its scientific nature, or transform it into an AI company that makes products. The company was already having problems Balance between scientific research and past product development.

And DeepMind is not alone. OpenAI, DeepMind’s implicit rival, was faced a similar identity crisis that is evolving from an AI research lab to a Microsoft-backed for-profit company rents out its deep learning models.

While DeepMind doesn’t have to worry about its unprofitable research just yet, and is increasingly entangled in its owner’s business dynamics, it should think deeply about its future and the future of scientific AI research.

This article was originally published by Ben Dickson on TechTalks, a publication that examines technology trends, how they affect the way we live and do business, and what problems they solve. But we also discuss the evil side of technology, the darker effects of the new technology, and what to look out for. You can read the original article here.

Published on January 12, 2021 – 09:54 UTC

Like it? Share with your friends!



Your email address will not be published. Required fields are marked *