The Office of the President of the US has released its long-anticpated report on AI called “Preparing for the Future of Artificial Intelligence.”
Here is our summary of it, sprinkled with our thoughts.
One of the promises for AI is that it has benefits for society as a whole. The Report notes, for example, that Veterans Affairs is using AI to predict medical complications and treat combat wounds, leading to better healing and lower medical costs.
John Hopkins University is also using AI to predict complications to reduce infections at hospitals.
In the future, AI will be used to review electronic health records to analyze health risks but, in the view of the Digital Finance Institute, such changes are fraught with issues of privacy law and consent, as well as insurability implications and represent a peril of AI and automation.
In infrastructure, AI can be applied to improve flows of traffic with AI applied to smart traffic light systems that could save the US economy $121 billion per year and eliminate 56 billion pounds of CO2 annually. For air travel delays, if air traffic systems were improved with AI and machine learning algorithms to use airspace more efficiently, it could save the US economy $50 billion per year in air transportation delay costs.
Cars equipped with AI that can self-drive also hold enormous promise. The benefits of self-driving cars include 90% fewer accidents, 40% lower insurance costs, elimination of drunk and drugged drivers and the ability of people with disabilities to have new-found mobility.
There is some thought being given to leveraging AI to improve animal migration, especially of endangered species using AI image classification software to analyze tourist photos from social media sites to identify animals and build a database of their migration.
We believe, at the Digital Finance Institute, that it could work for whales and at risk populations of lions in Africa, for example, or polar bears in Canada’s arctic. Such tracking and tracing can protect Big Game from poachers and save endangered species.
In the same vein, autonomous ocean vessels with AI can be used to detect and report illegal fishing in oceans and other illegal activities such as drug trafficking off the coast of British Columbia.
In the global scene of things, at the Digital Finance Institute, we believe that AI has the potential to make a material impact on making humanitarian aid delivery safer and more effective and to provide personnel management, navigation, communications and medical assistance in a more tailored fashion and more efficiently, saving billions annually. Some similar initiatives the Report notes that Americans are looking at include machine learning to detect where poverty zones are likely to develop to assess where assistance is needed.
The US government is supporting the AI sector in a number of leadership ways, such as ensuring that it is one of the first customers of private sector development. Other support includes investment, supporting challenges for students in the space and creating policy and legal environments that allow innovation to flourish while protecting the public.
The issue of diversity in AI is problematic. In the US, just 18% of computer science graduates are women, down from 37% in 1984. In order to solve the problem, the US is assisting with the support for diversity in AI.
The diversity arises in the coding of AI and also in the issue of the fact that AI has been used to create bias predictive results, calling into question issue of justice and bias generally.
Some of the key takeaways of the Report include helpful working definitions.
There remains no accepted definition of AI but despite this, there is agreement on the distinction between narrow AI and general AI. Narrow AI involves the use of AI for narrow purposes such as names, language, image recognition, banking or self-driving cars. General AI, on the other hand, involves AI that exhibits intelligence behavior across several cognitive tasks equal to a human. General AI will not be available for decades.
Machine learning, an application of narrow AI, involves a statistical process that starts with a body of data and derives a rule that explains the data and can predict future results. Machine learning from data relies upon statistical methods, as opposed to AI that involves coding rules into software for decisions. Machine learning does not solve problems, per se, but rather it finds solutions for problems based on data provided. Often the combination of machine learning and human intervention produce better results – cancer radiology results, for example, reduce errors 85%, a result that is better than either one on its own. It goes to show the importance of providing intelligence capital in the AI space.
Deep learning uses units of input values to produce an output which is used to build upon other units and so on until the result programmed is achieved.
Autonomy refers to the ability of system to operate and adapt to changes without human intervention, such as an autonomous cars, financial trading and security systems that identify and correct vulnerabilities.
Automation refers to machines performing work of humans and does not necessarily involve AI except that AI will automate jobs and cause unemployment.