Ditto AI

What we can learn about AI from developments at Google's DeepMind

Posted by Ditto on Mar 20, 2019 3:55:44 PM

It’s no secret that DeepMind has long been at the forefront of developments in AI – they’re famous for pushing the technology to its limits and finding out just how much it can actually do.

DeepMind are mostly known for their research and experimental applications of AI, though. They’re not best known for demonstrating the business benefits of AI – they’re the ones making those possible ten years from now.  

It pays to keep up with DeepMind and their latest projects, though, because those experiments will likely impact the future of AI in business and the breadth of its potential uses. We’ve taken a look at two of their recent high-profile developments to get you up to speed.


1. AlphaStar

The most recent breakthrough celebrated by the DeepMind team was the success of their ‘AlphaStar’ AI in beating a professional Starcraft II player. Starcraft II, a competitive ‘real-time strategy’ game, may seem like an unusual forum for DeepMind to be focusing their efforts on, but its success against a human player reveals a lot about the state of their artificial intelligence. 

What to look out for

Starcraft II is considered a ‘grand challenge’ for AI, because it requires flexibility of its players. What AlphaStar’s success tells us is that AI has the potential to adapt to real-time challenges, and to master several types of knowledge that could have a significant ripple effect in the world of business strategy.

According to DeepMind, the AI had to master and work with:

  • Game theory
  • Imperfect information
  • Long-term planning
  • Real-time developments
  • A large ‘action space’ (hundreds of variables at once, in this case the ‘units’ and buildings controlled by the player and their actions).

What at first glance is a seemingly irrelevant accomplishment for AI is in fact a sign of its progress, and the direction that DeepMind, at least, will be taking it in the future. The technology’s potential to ‘live-strategise’ using variables beyond raw factory data could mean that the ROI of AI is set to make an unprecedented leap forward. Flexible AI is valuable AI.

AI can increase bank's revenues by as much as 30 percent text-only CTA

2. AlphaZero

AlphaZero is the latest iteration of AlphaGo, the system that made headlines for beating a champion Go player in 2016. Again, though, what does beating humans at board games have to do with business?

What to look out for

The value in AlphaZero lies less with the final results of its robot vs. human clashes, and more with how it gets good enough to win. Previous board game winning AI has required input from professional players to ‘teach’ it how to succeed.  

AlphaZero teaches itself. Supplied only with the rules of the game – whether it be chess, shogi or Go, it plays millions of games against itself and learns through trial and error. DeepMind calls this process ‘reinforcement learning’, and it’s extremely effective – it took AlphaZero nine hours to master chess.

What does this mean if you’re someone investing in AI for business? The applications are manifold, but consider the potential of the technology as it is able to process more complex sets of rules – those of law, business strategy or medicine, for example. In collaboration with human input, the technology behind AlphaZero represents a leap forward in AI’s capability.  

AI in practice: developments in diagnosis

It's useful to think about potential outcomes for this exciting technology, but what does it look like when it's applied to a real-world problem? We've been given an example with DeepMind's recent foray into the world of medical diagnosis. 

After working with specialists at London's Moorfield Eye Hospital over the past three years, DeepMind has announced that they are prototyping a product that promises to - in the words of one of the doctors involved - 'fundamentally transform' opthamology. The technology can accurately diagnose over fifty eye disorders, including diabetic retinopathy. 

The algorithm, which is able to match (and exceed) the accuracy of eye specialists with decades of experience, does more than just signpost the state of AI in 2019. It also highlights the priorities of AI developers going forward, as DeepMind have made it clear that 'it will not just offer diagnoses, but also be able to explain how it arrived at the conclusion, and how certain it is of the result.' Explainable AI, then, is at the forefront of the latest developments in the field, and a necessary accompaniment as the technology begins to match - and advance beyond - human capability. 

Clearing the path

The sorts of advances being made by organisations like Google’s DeepMind point to AI’s future as a key agent in making informed business decisions alongside its human counterparts. There are hurdles on the path to this future, though – and one of the most significant involves transparency. Over 80 percent of the UK still wouldn’t trust artificial intelligence with their finances, despite reports from PwC last year that AI could improve the global economy to the tune of 15 trillion dollars by 2030.

Despite AI’s rapid advancement, transparency remains an issue. If we’re to make anything of DeepMind’s achievements, explainable AI has to become the standard – especially where its business benefits are concerned. We've now got a concrete example of that with DeepMind's entry into the medical market. If artificial intelligence can explain its strategic decisions, its potential will move beyond the chess board.

AI can increase banks' revenues by as much as 30 percent wide CTA

Topics: AI, Explainable AI, Latest Developments, Future of AI