The age of artificial intelligence is well under way.
The concept of general AI, an advancement that would enable the tech to effectively surpass human abilities in nearly every task, is still decades (or more) off. But companies across corporate America are deploying the technology in operations ranging from supply chain, legal, customer service, recruitment, and IT. And not just as test projects, but as major enterprise-wide initiatives that are leading to millions of dollars in savings.
And while industry and experts have long-pushed for a set of standards to guide the technology, federal action is lagging well beyond the pace of adoption. But the rapid uptick among enterprises is putting new urgency on providers like Google, Microsoft, and IBM to more forcefully call for industry standards around AI and outline the protocols they have in place to ensure the services they provide are ethical and trustworthy.
The adoption of the tech has "moved beyond this science experiment world — or doing a lot of proof of concepts or pilots internally — into a space where companies are now actually really getting ready to or are already deploying AI into production systems," said Tracy Frey, the director of product strategy and operations for AI at Google Cloud. "It really brings to the fore these questions of responsibility, if they haven't already been raised."
Take facial recognition, for example. As headlines swirled around the potential issues with the technology, IBM put a halt on the development of its own software. Microsoft was even advocating on the issue as far back as 2018, when one of the company's top executives publicly called for federal regulation of the technology.
Read more: One of the leading GOP voices on AI says the party's priority is on protecting consumers from the dangers of the technology, while also making sure America doesn't lose the innovation battle with China through over-regulation
The stakes are huge — particularly as the coronavirus pandemic accelerates the pivot to digital. Services like AI to help retailers improve the shopping experience can be a major selling point in convincing customers to also invest in their cloud services — an over $700 billion market that is typically branded as a battle for dominance between the tech giants.
There's been "two years worth of digital transformation in two months, so there is some urgency that comes with that," said Microsoft chief responsible AI officer Natasha Crampton.
AI is big business, so the stakes are high
Many of the largest industry players have already had public issues where their AI systems have produced unpleasant results.
Google, for example, was forced to publicly apologize in April after one of its AI-based technologies identified a thermometer in the hands of a Black person as a "gun," versus labeling the product in a similar image with a white person as an "electronic device." A conversational AI tool that Microsoft deployed in 2016 quickly turned racist, Amazon scrapped a recruiting tool in 2018 that showed bias against women, and Facebook's ad algorithms have been criticized for discriminating based on both gender and race.
AI systems that aren't properly tested for bias and fairness — that aren't checked for being "ethical" — could have seriously negative real-word consequences. An oft-cited example is the potential ramifications of an AI-backed machine that determines loan eligibility discriminating on the basis of race (one study suggests that this has already happened).
To fight against this kind of algorithmic bias, IBM, Microsoft, and Google are working to have audits in place to make sure this kind of thing doesn't happen. All three have invested heavily in operations dedicated to the trustworthiness of AI, including propping up new business units and cross-functional advisory committees, appointing new leaders, establishing mandatory training programs, and creating procedures to review the algorithms themselves. (Notably, Google faced controversy for its external AI ethics board last year and disbanded it in a matter of weeks.)
IBM is even eyeing this area as a revenue stream: Offering up toolkits to help developers verify metrics like fairness in AI algorithms, which is a centerpiece to one of IBM's core products: Watson OpenScale, a platform that monitors for adverse events in models.
Determining success and tapping the ecosystem
Judging the success of something as nebulous as trust in a specific type of technology is hard.
Part of the challenge, according to Frey, is understanding how different societal biases would manifest themselves within AI applications. But understanding those biases is critical to safeguarding the data that is used to power the AI applications, Frey said — an argument that many other experts make, given the necessity for robust information sets to train the algorithms to act appropriately.
"The question is no longer is this actually likely to happen," she added. Now, there is an "understanding of what the experiences are across the full range of diversity, for example. And so knowing that, what do we want to do here in order to effect positive change?"
But there is a limit to just how much the companies can accomplish on their own — particularly as AI startups continue to draw major funding from venture capitalists and other investors. That's where the ecosystem comes in, with partners like Team Gleason that are helping Microsoft create an open dataset to help power software to help those suffering from ALS.
And it's also a reason why the companies are so adamant about the role the federal government must also play — like requiring that developers like Microsoft "make available the technology in a way that it can be tested by third parties," per Crampton.
"We increasingly understand what these challenges are, but we don't always have easy to implement ways of mitigating them," she added. "There is much to be gained by working together on some of these hard problems as opposed to competing on them."