AI research lab DeepMind’s reported attempt to get more autonomy from its parent company Alphabet has fallen flat. As per a recent Wall Street Journal report, DeepMind’s team has been negotiating with the tech giant for two years now.
Founded in 2010, DeepMind was bought by Google in 2014. DeepMind has been behind some of the most path breaking innovations in recent years, including AlphaFold, a solution to the 50-year-old protein folding problem.
Google-DeepMind Deal
DeepMind founders have argued that the company must have the same legal structure as a non-profit, reasoning that a single corporate entity should not control powerful AI innovations and discoveries. Google, which shelled out $600 million for acquiring DeepMind, snubbed the proposal.
Google has been increasingly pushing DeepMind for the commercialisation of its work. In the first year after the acquisition, DeepMind reported zero revenue. However, that soon changed when DeepMind published a paper on WaveNet, a new approach for speech generation. Currently, WaveNet is deployed in Google Assistant. DeepMind has also come up with an Adaptive Battery feature for Android devices and an AI system to reduce energy costs and cool data centres.
In an earlier interview, when Google CEO Sundar Pichai was asked about DeepMind’s role as a financial driver for the company, he said, “Looking at the pace of progress, I think we will have AI in a form in which it benefits a lot of users in the coming years, but I still think it’s early days, and there’s a long-term investment for us.”
Apart from the commercialisation bid, the ethical oversight board is another bone of contention. At the time of acquisition, Google had promised it would set up an ethics board to ensure fair deployment of DeepMind’s technology. In 2017, DeepMind did get an ethics board, however, details about the board including its members have been largely kept under wraps. As per the agreement, if DeepMind succeeds in its core mission of building artificial general intelligence (AGI), the holy grail of AI, the control of this tech would lie with this board, an Economist report stated.
In a tight squeeze
DeepMind has not made a profit till date. Though it doubled its revenue from £103 million to £266 million, the expenses increased from £568 million to £717 million in 2019. DeepMind clocked an overall loss of £477 million for the same period.
DeepMind is heavily dependent on Google, which is also its biggest customer. Further, DeepMind’s infrastructure runs mostly on Google’s cloud services and TPUs. The former’s main area of research is deep reinforcement learning that requires expensive compute resources.
Limitations
Commercial AI projects provide end-to-end and ready to use solutions. On the downside, such projects are not open to modification. In a few cases, the owners of the tech also claim ownership of the data generated when you use their AI systems.
OpenAI drew flak when it decided to give the exclusive license of GPT-3 to Microsoft. Tech populists like Elon Musk panned this move. “OpenAI should be renamed ClosedAI — for all intents and purposes they are a for-profit company,” said Oren Etzioni, CEO, Allen Institute for Artificial Intelligence (AI2).
Autonomy is paramount in research to avoid the concentration of power among tech oligarchies. Following the unceremonious exit of two of its Ethical AI researchers in the past few months, Google is already in hot water. Further, the latest decision to not allow larger autonomy to DeepMind is widely perceived as a wrong move.
