Can We Really Trust Artificial Intelligence?

While calculation is not consciousness, my continuing interest in artificial intelligence leads me to wonder just what computer algorithms can do in terms of interpreting what is known today as “big data” – and where it might lead us.

My interest in the potential consequences of machines evaluating the information in other machines came to light a few weeks ago when I travelled out of state. While getting gas the pump malfunctioned and I moved to another pump to finish fuelling.

Then my credit card was turned down and I had to call the company. In addition, a possible “fraud alert” came in to my email. Since the pump malfunction had been odd I thought maybe I had been hacked, but when I called the company it turned out that there was no fraud at all – the “computers” had just decided that my anomaly was irregular enough to warrant inconveniencing me by shutting off my credit.

When I complained about this the fraud person said simply, “It’s our policy. If the software triggers an alert we have to act.” It reminded me of Treasury Secretary Hank Paulson’s famous comment when asked why the banks needed an $800 billion bailout in 2007.

He said, “The computers told us.”

The problem is that much of this “artificial intelligence” is unfounded, unproven, and just plain wrong. Just as there had been no fraud on my credit card, just a glitch at a gas pump – but how do you hold a computer program accountable?

https://www.youtube.com/watch?v=F-YbZdaALHE

Companies like MetaMind specialize in what is known as “Deep Learning” – using software algorithms to analyze the data that comes in.

Says Bloomberg Business News:

“The company has put free A.I. online for two reasons, [founder] Socher says: to attract clients for its customization business and to feed data to its servers so the software can keep learning. 'As we explore and observe people using the platform, I think the platform will get smarter and smarter,' he says.”

Here the program provides a score for how closely related two sentences are:

One of my early forays into this field of data analysis led me to the latest in “search” – and I discovered a startup called Twine – touting the future of the “Semantic Web” – so I was wondering what its CEO, Nova Spivak, is up to these days.

His new venture, Bottlenose, specializes in “Trend Intelligence” – or analyzing what is capturing the bulk of attention on social sites like Facebook and Twitter. From the site:

"[We can help] discover developing trends and get early-warnings about breaking news and emerging hot topics. Track what’s trending in your industry and among your customers to gain first-mover advantage.”

“One of the top US Cable TV networks used Bottlenose to detect and discover breaking news stories around celebrities and entertainment. Bottlenose showed them all the topics, people, photos and stories that are emerging and trending in real-time.”

One of the main applications of these kinds of algorithms is to monitor social media for references to a client’s product or service and uncover either very positive or negative references; this is part of reputation management.

Mako – the program behind the video at the top of the article – is a subscription based software package that promises that you can trash your keyboard – it lets you control your computer with your voice. This is certainly not entirely new; Naturally Speaking from Dragon has accomplished similar tasks, but the introduction of AI to voice recognition puts us into a new paradigm.

Harvard Business Review published a piece on AI for business in March. One of the insights was that “an early analytics insight at Osco Pharmacy uncovered that people who bought beer also bought diapers. But because this insight was counter-intuitive and discovered by a machine, they didn’t do anything with it. But now companies have needs for greater productivity than human quants can address or fathom. They have models with 50,000 variables. These systems are moving from augmenting humans to automating decisions.”

The HBR article mentions Watson, the IBM initiative that created the Jeopardy champion of champions using amazing search and analytic technology; the main application for Watson seems to be in medicine. Will the software carry malpractice insurance?

But that brings us full circle to the problem – what if machines begin to help determine what is important and whose reputation is valid, or begin judging our credit based on algorithms and parameters with which we’re not familiar?

Returning to the financial crisis of 2007, much of the problem centered on badly evaluated loan applications. In that situation banks were apparently approving loans based on misinformation and poor criteria. The question then becomes, would computers have fared any better?

And who is ultimately responsible or to be held accountable for these decisions?

The end user? The programmer? God? One thing is for sure – attorneys will make a fortune sorting it all out.