Artificial intelligence is quickly being deployed into many areas of business and society, but experts and the companies that build the technology are urging policy-makers to be wary of it’s biggest weakness: the humans that design and shape it.
“The obligation that we all have as the producers and consumers of this technology is to take a step back and actually start to get a deeper understanding of some of the consequences that can happen,” said Norm Judah, chief technology officer for Microsoft Services, the global branch of the Seattle-based tech giant that handles business and consumer software-based offerings such as the Azure cloud platform, AI and Office365.
Judah said in an interview that rapid advances in cloud computing and algorithms have accelerated the technology to the point that policy is needed to address potential ethical conflicts.
“AI systems that learn are organic and continue to learn … but with lack of governance and guidance they can actually start to inherit and build biases,” he said.
Unlike automation, where systems are simply told to do specific repetitive tasks, the nature of AI is to find patterns, learn from experience and make decisions based on a large volume of data.
It is, however, often only as good as its systems and training data, which are designed by humans.
Microsoft Corp. recently self-published a 150-page book called the Future Computed: Artificial Intelligence and its Role in Society which outlines the many advantages of the technology, while also suggesting steps that must be taken to better protect society from its potential misuses.
For example, AI could be used to make employment decisions, but if the training data was based on past information then the system may find itself biased toward white, male candidates in industries historically dominated by that group.
Similarly, an algorithm weighted toward profitability could potentially be used to unfairly select who should get a mortgage, or whether or not someone should get medical treatment.
There are pitfalls for business as well, such as the possibility of inadvertent price fixing in an industry with few major players who are all using the same available data.
“There are always people involved and the role as humans, to a large degree, is going to be judgement and accountability,” Judah said, while adding that the ability to understand information and data will have to be a core competency for people in 10 years because of AI’s fast growth.
“AI deals with probability and inexactness … most people are used to dealing with yes or no, black or white, and many will simply determine 72 per cent is ‘yes’ but forget that there is a 28 per cent possibility of ‘no’.”
Organizations, including those in the public sector, need a manifesto for AI that outlines what they believe and their principles, Judah said. That way when the hard questions are asked and ethics come into question, there are guidelines in place.
That could simply mean how user data is stored and used when it comes to privacy and trust. The discussions can also include policies and discussions on ensuring denial of service based on race or income level doesn’t occur.
“If you actually ask these kind of questions, whatever your answer comes out to be there should be actions that are taken to mitigate that in some way in the design of an overall system,” Judah said.
Judah is visiting Toronto on Thursday for Future Now, an AI-focused event hosted by Microsoft for industry leaders to discuss the technology and how to be responsible when carving out its future.
Michael Karlin, an adviser at the Treasury Board of Canada Secretariat, will also be a speaker at the event. He specializes in the public policy considerations of AI and has been working with the federal government in finding ways to use the technology.
“I’m generally positive on the potential for AI to do good for society. We humans have blind spots, biases and mental shortcuts that we take all of the time and this is a different type of intelligence to give us new insights,” Karlin said, adding that bias in AI’s decisions and recommendations are also a concern for him.
“What keeps me up at night is not knowing what is going on in the background.”
Canada is quickly becoming a hotbed for AI startups, research and investments from major tech giants such as Google, Microsoft, Apple, Amazon and IBM. Meanwhile, the federal government has been testing low-risk AI tools in situations such as helping Canadian Heritage choose which communities would receive micro-grants, Karlin said.
“We are very cognizant that there are some of these issues around data biases that the private sector deals with and we care a lot that the systems we put forward are ethical and responsible,” he said.
“In terms of biases, the first step to solving that is to not deny it exists.”
Artificial intelligence has already had a positive impact for millions of people across various sectors and experts say it is only a matter of time before the technology will be used in most facets of society.
Both Judah and Karlin say the key is to embrace the coming changes and for government and companies to work together on shaping policies to protect everyone.
“There will be societies that just define a set of norms of expected behaviour and capabilities and then companies can decide how they want to operate as part of those norms,” Judah said. “I feel that it is a partnership.”