Daily Archives: November 29, 2019

Fresh efforts at Google to understand why an AI system says yes or no launches Explainable AI product

Posted on by

Google has announced a new Explainable AI feature for its cloud platform, which provides more information about the features that cause an AI prediction to come up with its results.

Artificial neural networks, which are used by many of today’s machine learning and AI systems, are modelled to some extent on biological brains. One of the challenges with these systems is that as they have become larger and more complex, it has also become harder to see the exact reasons for specific predictions. Google’s white paper on the subject refers to “loss of debuggability and transparency”.

The uncertainty this introduces has serious consequences. It can disguise spurious correlations, where the system picks on an irrelevant or unintended feature in the training data. It also makes it hard to fix AI bias, where predictions are made based on features that are ethically unacceptable.

AI Explainability has not been invented by Google but is widely researched. The challenge is how to present the workings of an AI system in a form which is easily intelligible.

Google has come up with a set of three tools under this heading of “AI Explainability” that may help. The first and perhaps most important is AI Explanations, which lists features detected by the AI along with an attribution score showing how much each feature affected the prediction. In an example from the docs, a neural network predicts the duration of a bike ride based on weather data and previous ride information. The tool shows factors like temperature, day of week and start time, scored to show their influence on the prediction.

Scored attributions shows by the AI Explainability tool

Scored attributions shown by the AI Explainability tool

In the case of images, an overlay shows which parts of the picture were the main factors in the classification of the image content.

There is also a What-If tool that lets you test model performance if you manipulate individual attributes, and a continuous evaluation tool that feeds sample results to human reviewers on a schedule to assist monitoring of results.

AI Explainability is useful for evaluating almost any model and near-essential for detecting bias, which Google considers part of its approach to responsible AI.

Source: Explain yourself, mister: Fresh efforts at Google to understand why an AI system says yes or no • The Register

Internet Society CEO: Most people don’t care about the .org sell-off. Grabbing money at the expense of non-profits is fine by everyone we didn’t consult or listen to their opinion.

Posted on by

El Reg has quizzed Andrew Sullivan, the president and CEO of the Internet Society (ISOC), about his organistion’s decision to sell the non-profit .org registry to private equity outfit Ethos Capital.

We have previously covered the controversy over the proposed sale, the continued failure of ISOC and DNS overseer ICANN to answer detailed questions, and efforts by both to push the deal forward even while opposition to it grows.

Your correspondant asked Sullivan whether he expected the amount of criticism from the internet community that has erupted in recent days.

“I did expect some people to be unhappy with the decision, I expected some pushback,” he told The Register, adding: “But the level of pushback has been very strong.”

He was aware, he says, that people would not like two key aspects of the decision: the move from a non-profit model to a for-profit one; and the lack of consultation. He had explanations ready for both: “The registry business is still a business, and this represented a really big opportunity, and one that is good for PIR [Public Interest Registry].”

As for the lack of consultation: “We didn’t go looking for this. If we had done that [consulted publicly about the sale .org], the opportunity would have been lost. If we had done it in public, it would have created a lot of uncertainty without any benefit.”

Overblown

But when we pressed him on the fact that the concerns seem much deeper and broader than that – one ISOC Chapter has accused the organization of “severely harming” its reputation “by even contemplating this transaction” – he rejected the idea.

“I think claims that there has been an outpouring of support against the sale are overblown. If you look there is a relatively small number of people complaining. We may be overstating the feeling; most people haven’t noticed. Most people don’t care one way or another.”

It’s hard to simultaneously argue that there was no need for consultation and then claim that the lack of responses indicates implicit approval, we note. More importantly, though, what about the 10 million registrants of .org, the vast majority of which are unlikely to hear about the sale at all and who likely bought their .org domain precisely because it represented a non-profit ethos?

Source: Internet Society CEO: Most people don’t care about the .org sell-off – and nothing short of a court order will stop it • The Register