Will the Next Generation Have Artificially Intelligent Assistants?
Artificial intelligence (AI) describes projects involving purpose-driven computer hardware and software. When aimed at economically valuable goals, AI projects have proven to be helpful, but they are expensive to develop and are not yet up to some tasks. A big part of the challenge is overcoming shortcomings in how we teach computers to “think.” Progress is expected to continue at a rapid pace, producing capabilities that overwhelm our readiness to handle social disruptions.
In raw computing power, computers are superior to humans, but equipping the computer to digest language, context and intent is a huge challenge. For example, how do humans know that a rose is a plant and a tree is a plant, but a tree is not a rose? For adult humans, this seems automatic, but teaching a computer skill with these categorizations is not so automatic.
AI Computers appear wizardly in structured contexts that require brute force logic and calculations to reach the goal. AI computers are smart enough to excel at chess, (IBM’s Deep Blue), recently won at Texas Hold ’em Poker (Deep Stack), and in the game of Go. Aside from the rules of the game and the location of exposed game pieces or cards, the AI system must derive the opponents’ unstated strategies which sometimes change.
In the law profession, tedious tasks such as document review are usually relegated to junior lawyers, but even at their low hourly rate, document review can incur huge billable charges. There are many AI systems that rely on natural language processing to perform a legal document search. A more advanced AI tool set is Lex Machina, a Lexis Nexis service that can analyze courtroom strategy suitable for handicapping the judge, topic and presenting attorney. These assessments can predict whether it makes more sense to continue through the court process or settle. There remains plenty of work for human lawyers, since just one-quarter of a typical lawyer’s work can be handled by today’s AI.
AI computers have an affinity for parsing structured data, information that fits in categories such as billing codes or lab test values. Other data that is less structured remains impervious. Symptoms, treatments, images, reports and notes that doctors write about each patient are needed to complete the picture leading to diagnosis, preferred treatment and prognosis. Computers are less capable with unstructured data because digesting it requires making inferences and an understanding of context and intent, something that computers have not yet mastered.
In Japan, the use of care-bots is successful and beginning to flourish. These robots have limited intelligence, but perform duties as game-players or substitutes for live “comfort” animals. The care-bots work well with elderly patients and those with dementia. AI systems with similar depth include the Alexa intelligent personal assistant. It and its competitors (Siri and Cortana) are intelligent in the limited sense that they do natural language processing and pass the results to a search engine such as Google or Bing.
AI systems with deeper intelligence are in use by US physicians for offering help with diagnosing and prescribing. Wary of the time and expense required for FDA approval, companies engineering medical AI systems – at least for now – are careful not to describe them as diagnostic tools but rather as “information banks.”
Already there is wide variation in the smarts of AI systems. In some instances, the AI device is rudimentary such as a care-bot or Alexa. In other cases, the AI device supplants some functions of human professionals (in the practice of law or medicine).
The bigger question is, when will AI systems feel like genuine human personal assistants? The answer is difficult because a well-rounded AI assistant would require strong language skills, general knowledge and depth in one or two areas that match the owner’s knowledge, and soft skills such as empathy, humor and rudimentary human morality. Much of that would come from software and some from the raw horsepower of computer chips.
A rough guess at the birth-date for the first AI personal assistant is suggested in this anecdote: Mark Papermaster, CTO at AMD recalled a project where 82,944 computers were connected to mimic the actions of a human mind. It worked well enough, but took 2400 times longer to produce answers than a natural human would. That implies that the horsepower of 198,000,000 computers would be needed to keep pace with a human mind.
Moore’s Law describes the evolutionary track for computer power, i.e. the number of transistors in a PC doubles every 2 years. In some periods, Moore’s law understates the speed gains. If we assume that Moore’s Law can apply, it will take 27 to 28 years for computing technology to evolve a PC with the equivalent power of a human’s smarts. Adding actuators and a plastic body would complete a humanoid robot. The whole project could be done in just one human generation.
Indeed, the option of having a loyal, thoughtful robot friend could make radical changes in how society evolves. We need to invest some thought and discussion into how that may alter our lives and what safeguards we would want engineered into those “companions.”