When Google’s artificial intelligence program “AlphaGo” beat South Korean Go champion Lee Se-Dol in a 4-1 victory last March, many hailed it as a watershed moment for artificial intelligence.
The victory suggested that AI programs could strategize about beneficial and detrimental actions at a much higher level than previously achieved, a feat necessary for everything from self-driving cars to the old controversial idea of robots replacing human workers. But the real question at hand is, is this really some kind of general intelligence or just masterful idiot savants? The idea of expert systems is nothing new to AI, but are they a one-off solution that can play a masterful game of chess or Jeopardy but cannot be directly repurposed to play tic-tac-tie or Monopoly?
As astounding as Google’s achievement was, for the most part commercial AI has continued to plug away at virtually the same tasks it has been doing since 2014 – crunching large amounts of data, albeit at ever faster rates. Perhaps it’s simply the demand. According to a 2017 study by Forrester Consulting on behalf of Emarsys, analytics remains the most sought-after feature of business AI. In addition, research by Tata Consultancy Services found that IT and Marketing remain the dominant areas where companies use AI, the former monitoring machine-to-machine interactions and the latter monitoring and tailoring ad placement.
But perhaps it’s a deeper, industry-wide recognition that most of the current AI offerings on the market have substantial limits. After all, the machine learning and big data based AI that currently pervade are powerful tools for identifying associations in large quantities of data, but don’t have much on humans in terms of working out the complex phenomena of cause and effect, or to identify modifiable factors that can engender desired outcomes.
The failings of big data were on prime display with the spectacular failure of Google Flu Trends in 2013. The idea seemed sound – researchers from Google published a paper in 2008 detailing how they could use search data tuned to flu-related topics to produce estimates of flu prevalence two weeks earlier than the CDC. Just as a spike in search terms and Twitter hashtags can signal what toys are trending during the holidays, the researchers believed that they could use spikes in flu related search terms to monitor the spread of the flu. But then the algorithm missed the 2013 flu season by 140 percent.
What went wrong? Possibly the presence of confounding variables. As big data and machine learning powered AI’s gains in processing power, they can incorporate into their algorithms more and more information, more and more variables that may affect data associations. But with little human intervention, inevitably some variables may display strong correlation by pure chance, with little actual predictive effect. As Wired.com reported, Google’s algorithm was vulnerable to overfitting to seasonal terms that happened around traditional flu seasons, but were unrelated to the flu itself.
Even when models can accurately predict outcomes, they are still far away from providing any actionable suggestions toward alternate outcomes. This is especially pressing for fields like healthcare, where long term labor shortage and an aging boomer population has led to an increased need for more reliable and automated solutions. Dr. Jonathan Chen and Dr. Steven Asch describe the issue of current AI in healthcare in their 2017 paper on Machine Learning and Prediction in Medicine.
” Machine-learning approaches are powered by identification of strong, but theory-free, associations in the data...Models accurately predict that a patient with heart failure, coronary artery disease, and renal failure is at high risk for postsurgical complications, but they offer no opportunity for reducing that risk (other than forgoing the surgery)...The last mile of clinical implementation thus ends up being the far more critical task of predicting events early enough for a relevant intervention to influence care decisions and outcomes.”
So how does AI make the leap to the next step? How do we take the higher order, predictive and adaptive thinking Google demonstrated in its Go playing robot and bring it to bear on areas like healthcare?
The first step is to enhance the big data and machine learning with another layer of AI functionality – that of cognitive intelligence. What is missing in the AI learning simply from crunching the data at hand is an AI that can reason through and incorporate missing information, reason abstractly, plan, and problem solve. Mapping out scenarios and searching for potential outcomes to weed out confounders or even fill in missing data gaps through hypothetical scenarios. In real world application, it is substantially more complicated that reasoning through an opponent’s potential moves in a complicated game and altering game play to reach a desired outcome. But the foundation is there, and the technology is certainly there as well.
While machine learning and big data has dominated for years, a new breed of commercial AI has begun to emerge to meet the more complex demands of modern human machine interactions. The aim of cognitive intelligence AI platforms isn’t to supplant big data or machine learning. Rather, it’s there as a supervisor of sorts, monitoring how traditional AI process data, filling in gaps and identifying misinterpretations. Ultimately, the goal of a cognitive AI platform is to be able to complete tasks without the need for human supervision, to be able to quickly process unexpected or unfamiliar external input and adjust its response accordingly. To, as doctors Chen and Ash put it, predict events early enough to provide relevant intervention that increases the odds of a desired outcome. Whether that is better post-surgical results, predicting when flu seasons will hit to provide people with early warnings, or conducting genomic research in immunotherapy is all in the application. Wherever it is used, cognitive intelligence will be the true watershed AI, board game champions notwithstanding.
AJ Abdallat is CEO of Beyond Limits, an artificial intelligence and cognitive computing company that is transforming proven space and defense technology from NASA and the U.S. Department of Defense into innovative solutions to address large and emerging markets.
The views, opinions and positions expressed within these guest posts are those of the author alone and do not represent those of Becker's Hospital Review/Becker's Healthcare. The accuracy, completeness and validity of any statements made within this article are not guaranteed. We accept no liability for any errors, omissions or representations. The copyright of this content belongs to the author and any liability with regards to infringement of intellectual property rights remains with them.