Microsoft, Oracles, and Biased Ais

Solving Unfairness in AI Algorithms

Yesterday, the MIT Technology Review broke the news that Microsoft’s working on a tool to solve bias in AI’s foundational algorithms. Concretely, this means that they are working towards minimizing the fear related to trusting third-party Machine Learning products with sensitive customer data.

In addition to this fear of data leaks, there’s also the rational fear of unintentional bias in Machine Learning models.

Rich Caruna, one of the members of Microsoft’s team who is working on this project, was quoted as saying that since the AI industry is in its infancy, it is not feasible to assume that industry professionals have enough expertise to keep most bias out of their models with a high degree of accuracy.

What is Algorithmic Bias in ML and AI?

This is all well and good, but to understand the true importance of Microsoft’s project, one also needs to be able to easily conceptualize Algorithmic Bias.

Algorithmic Bias may be understood as the phenomenon in which AIs show similar biases to humans in the way that they make decisions. One contemporary example would be “Tay,” the now infamous chatbot from Microsoft, which debuted in 2016. The original idea behind Tay was to publicly showcase a system that could take in information and learn from what humans say online. Unfortunately, this backfired in a big way and Tay had to be shut down after becoming offensive in just about every way possible.

In the end, this was due to the fact that the complicated functions or Algorithms that ran Tay did not properly account for racist remarks and other offensive remarks that she could encounter online. Therefore, it seems that she took all of that in as well as if it were not only data to learn from, but also data to act on.

All in all, we could more precisely term this “classification bias,” which seems to be one of the types that Microsoft’s future tool aims to target. Given their experience with Tay, it might also reasonably be said that Microsoft is the perfect candidate for doing so.

In other words, they had a direct, detrimental experience related to classification bias and assuredly took in a wealth of data on it.

The Importance of Oracles

What this has come down to at this point is that Microsoft is building a “bias identifying oracle.” What you might not know yet is: what exactly is in an oracle, at least, in this context? The short answer is that an oracle is a continuous data feed put together with a verification service for that data feed.

In the context of the Crypto world, it finds real-world data and then verifies that data as accurate in order to release it to a Blockchain. It could be extrapolated, given this, that in the AI world, the only difference in what an oracle is, lies in substituting the words, “AI system,” for Blockchain.

Since Oracles have been claimed to effectively serve as catalysts for the execution of smart contracts, then it could also be said that they could block AIs from taking any action until certain data is verified.

Logically, since pthe roject is in its relative early stages, one could then surmise that Microsoft aims to use a combination of data feeds and something like smart contracts, to execute its tool. Saying this, however, is merely supposition based on the current usage of this technology in places the Blockchain industry, at this point.

Only the launch of a such a tool will truly determine if it prevents the occurrences of future, “Tays.” Lending credence to this possibility is the fact that oracles and therefore the AI systems with them, seem to be able to be secured by smart contracts.