AGI.. Still 50 Years Out?? Let’s See. Of course, a lot depends on how one defines it.
A CogWorld Custom Feature By Peter Fingar
Artificial General Intelligence (AGI) is often shrouded in confusion ... what exactly is it? Consider “intelligence” in general, meaning to have the ability to reason, use strategy, solve puzzles, and make judgments under uncertainty; represent knowledge, including commonsense knowledge; plan; learn; communicate in natural language; and integrate all these skills towards common goals. Estimates of the time needed before a truly flexible AGI is built vary from 10 years to over a century, but the consensus in the AGI research community seems to be that the timeline discussed by Ray Kurzweil in The Singularity is Near (i.e. between now and 2045) is plausible. A lot of the long-range estimates are based on the top-down theory of simulating the human brain via neuroscience and human cognition. We reached out to a man who is in the business of early forms of AGI in the here and now, based on a more bottom up approach. Here’s our report after talking with Mounir Shita.
“Everything just works.” That’s the vision Mounir Shita sees when artificial general intelligence is embedded in everyone’s connected devices: smartphones, Internet of Things, automobiles, medical and industrial tools, and more. The idea that everything around you just works is neither new nor exceptional; every consumer electronics company wants their gizmo to work right out of the box. Shita, on the other hand, wants all your gizmos to work together and proactively help you achieve your goals – and here’s the kicker – even when you’re not yet aware of what your goals are.
Shita is the founder and CEO of Kimera Systems, one of a number of hot new artificial intelligence startups. What makes Kimera Systems unique is its claim that it has developed something the vast majority of AI experts say is impossible for years, possibly decades to come: artificial general intelligence or AGI.
AGI is often defined as machine intelligence that can successfully perform any intellectual task that a human being can. To reach this goal, venture capitalists and technology giants like Google and Intel have poured billions of dollars into research and development to digitally recreate the human brain. Shita believes this approach is fundamentally flawed, which is why six decades of R&D into this effort – beginning with Alan Turing – has not made any significant progress, and why the experts believe it could be as long as 100 years before it becomes possible to build the very technology Kimera claims to have launched onto Android smartphones in a closed beta in August of this year.
“I have some problems with how the AI industry defines ‘intelligence,’” says Shita. “First, there really isn’t an accepted scientific definition. I ask AI scientists all the time how they define this term, and all too often they cannot really express it. But when they do, more often than not they come back to the brain, biology and neuroscience. The industry is stuck trying to digitize the human brain.”
Shita began his AI research in 2005 purposefully using a different approach: quantum mechanics. Shita argues that intelligence is a force exerted by all things over directional time, that affects all events and outcomes. As Shita defines it, intelligence (as it pertains to AGI) increases the likelihood of reaching a particular outcome by manipulating a chain of events to make that outcome possible. Using quantum mechanics instead of neuroscience as his divining rod, Shita successfully tested AGI use cases even before Kimera was formally founded in 2012.
From the beginning of his research, Shita along with his colleague Nigel Deighton - a telecom industry veteran based in the UK who joined up with Shita after serving as a vice president of research for Gartner – wanted to create a business case that was fulfilled by artificial intelligence, not a groundbreaking technology that would then seek a practical model. The pair spent countless hours examining the current global and digital economy, identifying what they believed needed to be “fixed” by their vision of a super-connected and intelligent future.
Shita and Deighton came up with a near endless number of use cases for their envisioned AI technology. Grounded by a desire to build a useful product, they pursued their R&D agreeing that the purpose of performing a task is not the task itself, but increasing the probability of achieving an envisioned future. Practicality, usability and scalability were all features required for their resulting technology.
To have AI work as they wanted, it was neither practical nor scalable to have to code a new algorithm for each use case. While it is impressive to build an artificial intelligence that can beat a grandmaster at chess, you could not use that same set of code to pilot a self-driving car, or visa versa. It had to be a single-algorithm at the core of their solution.
The challenges of a single algorithm are manifold, the biggest one being that it isn’t programmed to do anything. At least not anything specific. The algorithm must address everything, but the technology itself has to be taught, very much like a child learns.
“We had very interesting conversations with investors and prospective customers,” recalls Shita. “Everyone wanted to know what tasks our technology could do. But Kimera’s technology isn’t programmed to do any tasks, and conceiving of it in that way is very misleading. Complicating our story further, because we are starting with smartphone deployments most people we spoke to wanted to put us in the digital assistant box along with Siri, Google Now or Cortana. What Kimera has engineered is an infrastructure technology, not a personal assistant. In the future, with our technology in the network, the entire internet becomes your personal assistant.”
Paramount to Kimera’s product philosophy is iron-clad user privacy. User data is stored in “personal clouds,” and while these personal clouds can be hosted anywhere – by users’ network operators, a hosting company or even on users’ own hardware – the data stored within them are owned by the users themselves. Even Kimera does not have access to this data. While Nigel learns common sense by observing shared user behavior, the data is never shared between user accounts.
In 2013, as the company was preparing for its first public presentation of working use cases, Nigel Deighton suddenly and unexpectedly passed away. To keep his memory alive, the company named its technology after him. “Nigel” – the first human-like intelligence for connected devices – made its industry debut at CTIA 2013 in Las Vegas.
Not only does Shita believe that much of the AI research is proceeding down the wrong path by pursuing biology and neuroscience, he also thinks that businesses are not correctly conceptualizing the way AI should be used. “AI should not be about product enhancements,” declares Shita. AI certainly can be evolutionary, but it’s better to think of AI as revolutionary – don’t just make better products; change your business model.
Viewing AI as an enhancing technology is what Shita calls “the Blockbuster Trap.” Blockbuster Video was once the largest national chain of video rental stores. With the advent of the internet, the company offered its customers an improved website with a better way of finding stores and reserving video tapes and DVDs from their brick and mortar stores. Netflix, already disrupting the video rental industry with its DVDs-by-mail service, saw the internet as a medium for streaming video directly to end-users. Today Netflix is one of the biggest brands in entertainment and Blockbuster is all but forgotten.
Shita asks us to wonder what happens when the internet changes from a passive network to one that works proactively. This proactive global network is what Kimera is working to create, one which enables a multitude of devices, apps, content and services to proactively come together to assist a person in reaching their goals, whatever they happen to be. From a business perspective, Nigel seeks to enable products, services and information to find their own customers rather than the reverse.
Using mobile as the first of many major market segments he wants Kimera to address, Shita foresees the end of the app economy and the start of a new “hyper” peer-to-peer economy, where information and apps flow directly to devices at the exact moment that information is needed. This Nigel-enabled disruption would disintermediate the app stores and generate new ways for industries, business and individuals to create revenue. “I expect developers will embrace the ‘reverse app store model’ and personal cloud concepts to quickly cut out middlemen like app stores, social networks, and other content and service aggregators.”
Shita wants Kimera and Nigel to transform everything. “I want Nigel to be used to accomplish great things, and huge challenges. I want Nigel to solve global poverty and cure cancer. We are obviously a ways away from realizing that. But we will accomplish bigger things than most new companies right out of the gate.”
Does Shita really think Nigel will cure cancer? He does, and he’s serious about it. “I want my devices to understand that I have cancer and hypothesize ways to rid me of it. Nigel is not comprehending that I have cancer because it’s looking to cure me of cancer; Nigel is working to help me realize my goals and dreams, and cancer would stand in the way of me achieving them.”
The telecommunications industry appears to be warming to Shita’s point of view. It’s been nearly 10 years that mobile network operators in developed countries have been slugging it out in markets with greater than 100 percent mobile penetration. During this time, price has become one of the biggest competitive differentiators, forcing operators to race to thread bare margins. With the specter of 5G looming in 2020, requiring a new massive round of capital expenditures, operators are beginning to realize that their competitors aren’t really other providers in the market, but those companies profiting from their connectivity by delivering services and data over the top (OTT). Companies like Google, Facebook and Amazon.
OTT services exist because networks and devices lack intelligence. Kimera contends that Nigel fixes this flaw. Shita points to the recent news of AT&T acquiring Time Warner and Verizon acquiring Yahoo in an attempt to better compete against content aggregators. But Shita warns that they are merely seeking to become their own middlemen. “What they are missing is that Google and Facebook aren’t the problem, and becoming like them is only a band-aid to the problem. Owning your own band-aid is not a solution.”
By transforming passive connectivity into intelligent and proactive networks, Kimera’s Nigel stands to disrupt what has become the conventional OTT model, which the company estimates is easily valued at $2 trillion annually.
Shita says that Nigel will also change the conventional telecom revenue model of ARPU (average revenue per user) to ARPD – average revenue per device. Kimera states that Nigel “makes smart devices intelligent.” What it means is that there will soon be billions more IoT devices than there are mobile phones, and Nigel will integrate them all into a cohesive personal area networks where all devices work in concert to give a unified intelligence experience. For instance, sensors in fitness trackers, shoes and even underwear will actually make all those items intelligent medical devices. Not only could they together change user behavior on diet and exercise. If a person using these devices is in distress, they can summon the right help to the right location at the right time, potentially even before the person is him- or herself aware of an emergency medical issue.
Shita and his team are well aware that, while their theory is sound, it’s really proving the technology in the marketplace that will make the difference. In August, Nigel exhibited its very first production instance of learned behavior. Beta testers were told to go visit a movie theater. The Nigel technology, which learns unaided and solely from observing sensor data on the devices on which its installed, was able to first learn the concept of “movie theater” by using geolocation and publicly available information. It then observed that users were turning off the volume inside the theater. Within four days, Nigel “comprehended” that people commonly turn the volume off when they enter a movie theater, then began proactively doing that for all users, even if they had not yet taken their Nigel-enabled device into a theater. Kimera contends this is the first instance of common sense learning by a machine “in the wild.”
Kimera is working diligently to ready a major expansion of its closed beta to include thousands more users, and eventually take the beta test public in early 2017. Once they do, they are confident that the skeptics will become believers.
In the interim, the company is presenting its Theory of Intelligence to the AI industry. For example, the company was a major sponsor of the AI World conference in San Francisco this November, and was not only making several public presentations of both its technology and the business cases for it, they offered in-depth private sessions. “We have no illusions that people left our presentations believing we have achieved what so many experts say is not possible for decades to come,” acknowledges Shita. “But we do hope that people left thinking it’s possible we might have achieved real AGI. What we are saying is ‘here is our AGI and how it works. Come back in 2017 and we’ll show you the result of Kimera’s AGI running for 12 months."