The AI Lock In Loop
Much more awareness is needed about the importance of making the right kind of ethical decisions in artificial intelligence, according to Azeem Azhar, an entrepreneur, investor and tech industry veteran with a keen interest in AI.
He was referencing issues with Google’s racist imaging algorithm and the COMPAS recidivism algorithm “which is no better at predicting recidivism than random people.”
“It’s imperative that what we build are responsible AI systems.”
“My argument here is that AI systems are going to control the interface to resources. Whether that’s media, education, finance,” he said. “It’s imperative that what we build are responsible AI systems.”
It is an important issue to take into consideration when one thinks of the potential size of the AI industry. Some of the largest companies in the world right now are building artificial intelligence into their business models, said Azhar.
“Alphabet, Amazon and Facebook are really machine intelligence leaders and they have the application of machine intelligence in the core of their business models,” he said.
Azhar said that one of the reasons that these companies have been able to create so much value for their shareholders is something he has coined, “the AI lock-in loop.” He explained: “If you put some AI into a product, it will improve your product and it will create more usage, which will generate more data and that data can be used to improve your AI, which in turn improves your product, which in turn generates more usage and more data.” He said that, as a result, companies like Google have a really strong defence against new entrants to their markets.
According to Azhar, there is a lot of interest in artificial intelligence amongst CIOs and developers. He mentioned a Gartner survey of CIOs published at the end of 2017 which listed their priorities as AI and variations of it, such as “intelligent foundations”, “intelligent apps” and “conversational platforms”.
He also spoke of a survey of developers which found that 90% were intending to have “done something in AI” by the end of 2018.
This is no surprise considering the large volumes of investment from venture capitalists and large technology firms. Azhar also said that venture capital deals that are going on in the AI space are taking place in every sector, including those that one wouldn’t necessarily expect, such as real estate and law.
“Now we have tons and tons of deals going on there and that trend has continued,” he said. One research group looked at how enterprises are investing in cognitive and artificial intelligence technologies and calculated that $12 millions of investment was made into these tools, with 50% year-on-year growth expected through to 2021.
These high levels of investments may be a result of an extremely rapid rate of innovation, of which Azhar had first-hand experience. “Back in 2012, I ran a project to do some object detection for a product. We simply couldn’t get it to pass muster.
“Now the quality of object detection that is available almost for free by these cloud APIs is surpassing human performance.” He added that, in 1994, “we couldn’t beat chequers algorithmically,” but by December 2017 Deep Mind released Alpha Zero which took less than four hours to teach itself chess and beat the world’s best chess-playing program.
Azhar said that edge computing is on the rise because a lot of applications, healthcare, personal and wearable, are being built that will need to have inference and learning at the edge. Therefore, a lot of investment is going into silicon hardware.
“If I have a robo-surgeon operating on me, I don’t want it buffering."
“We can’t afford latency. With edge computing, the time cost of sending something to a data centre and having it come back is too much,” he said. “If I have a robo-surgeon operating on me, I don’t want it buffering at the moment it is about to suture up my skin.”
Having intelligence at the edge will require a new class of silicon which is why there are over 25 companies now developing new chips. “We haven’t seen this level of investment in silicon hardware for decades and that’s going to lead to what I think is going to be called a new enterprise stack,” he explained.
The analysis of the potential of the artificial intelligence industry led Azhar to pose some questions.
How do you articulate to the user how their data might or might not be used? How do you get informed consent?
How do you build chips that are powerful, yet low-power?
How do you manage software and firmware updates in this new edge-based world?
How are you going to manage cyber security?
He underlined the importance of managing cyber security with an invented nightmare scenario. Azhar said: “Imagine the IoT AI threat where you are held hostage by smart locks which are being controlled by the smart fridge and there’s no way of getting the firmware update to release you.”
You Might Also Read:
The World Needs More Computing Power:
China Intends To Be An AI Superpower:
Artificial Intelligence Needs Regulation:
New IoT Chips See, Think & Act Autonomously: