Banks will be able to use AI without explanation, but will not be able to attribute responsibility for errors to it

The group included representatives from the Bank of Israel, the Securities Market Supervisory Authority, the Capital, Insurance and Pensions Authority, the Anti-Monopoly Authority and the Ministry of Justice.

One of the key goals of the report was to solve the “black box problem” when working with deep learning technologies. We are talking about systems that can make accurate decisions based on huge amounts of data, but the decision-making process itself sometimes remains a mystery even to developers.

The group recommended not blocking the use of these technologies, but decided that liability for any damage caused to a client as a result of the work of artificial intelligence would be fully borne by the financial institution that used the AI. Banks and other institutional structures will not be able to shift responsibility for damage to the AI ​​platform developer or software provider. The inclusion of such transfer of responsibility in contracts with clients is strictly prohibited.

In addition, organizations will be required to ensure that the logic and operating principles of the system are “generally explainable”, but will not be required to explain to the client individually why a particular decision was made in his case. It also provided an exception for special cases with increased risk in the absence of adequate alternatives. This clause is intended to ensure that at critical stages for the client, where there is no human control, artificial intelligence does not become the “last resort”, not responsible to anyone.