It's neither artificial, nor intelligence. It's directive. It's binary code overlaid on a computer by a human. The variables the code incorporates are conjured from a human mind. The "intelligence" is conjured from without (the computer) not from within. That's "Artificial Intelligence."
AI is neither artificial nor intelligence, but it is quite useful in situations were the variables the code incorporates are inert, constant, or predictable.
Your fingerprint and face, though not inert, are constant. They are fodder for AI analysis. AI can capture your fingerprints and face and decide with reliable accurately to let you in or keep you out. Traffic patterns are predictable: Morning and late-afternoon traffic is predictably heaviest. AI can direct you (as we see in the Waze app) to alternative, less congested routes These are but a sliver of examples of AI's utility.
AI is considerably less utilitarian when the algorithm is coded to act in response to human action. We humans are sentient, self-direct beings. Our tastes and preferences are protean. We interact with one another in ways that are unpredictable. That said, we have established habits, our daily life can be a repetitive cycle. Beware of geting too comfortable, though. The habits and cycles can change instanter and permanently on a whim, an impulse, a coercion, a rumor, a fact, a feeling (and they do).
Because markets are outgrowths of human action, they are assured to be intractably vexing to AI-directed computers. Just when you think you have it figured out -- the right habits and established pattern variables are captured -- it all changes.
Perhaps you remember Long-Term Capital Management, a 1990s hedge fund run led by Wall Street's best and brightest: Nobel-winning economists and ultra-alpha traders. LTCM's founders were convinced they had figured it out. They went to work to exploit their genius.
Using quantitative models, the fund employed an arbitrage strategy to exploit deviations in value between liquid securities across different markets. The strategy was successful out of the gate: The fund returned (after fees) 21% in its first year, 43% in its second year, and 41% in its third year. The computers were successfully generating profits based on the empirically observed repeating patterns of human behavior captured and coded in the algorithms.
Success out of the gate, but a failure down the homestretch. The geniuses overlooked the fact that humans are never predictable in perpetuity. Patterns change, and often in the time required to bat an eye. Algorithms, on the other hand, don't. The financial markets were hit with the Asian financial crisis in 1997 and then by the Russian debt crisis in 1998. LTCM's algorithms failed to accurately capture the immediate human reaction to both events. LTCM lost $4.6 billion in less than four months in 1998. The whole shebang collapsed soon after.
Fast forward to the here and now and we find a lesser AI failure in the financial markets.
Zillow announced this week it will shutter its high-tech, house-flipping segment. The segment relied on algorithms coded to forecast house prices. The algorithms would formulate offers to buy houses today based on the prices it forecast to sell tomorrow. Zillow's algorithms were apparently predicated on human behavior that would adhere to the established trend in home prices, which was up. Bloomberg reported that Zillow even tweaked its algorithms to generate buy offers at higher prices at the moment so that it could sell houses at even higher prices in the future. Such was the established trend, such was the coding embedded in the algorithms.
To forecast that human behavior will change unpredictably and inconveniently is the consistently accurate forecast. It is also the forecast Zillow's algorithms ignored, because the magnitude of change (implied in market volatility) is impossible to codify (accurately).
Homebuyer behavior turned more discriminating and price sensitive. Homebuyers are no longer willing to unquestioningly accept a seller's price as the trend predicted. Zillow now finds itself holding a multi-billion-dollar residential portfolio unlikely to fetch the algorithms forecasted sales price. Zillow acknowledged the new reality (and the algorithms failure) with a $304 million write-down in the latest quarter.
Were ignorance and parsimony to blame?
No. As with LTCM, Zillow was privy to a wealth of data, an abundance of intelligence, and millions of dollars of cheap capital. The favorable ternary was still insufficient to guarantee success. “We’ve determined the unpredictability in forecasting home prices far exceeds what we anticipated and continuing to scale Zillow Offers would result in too much earnings and balance-sheet volatility,” conceded CEO Rich Barton.
We are no different than Zillow or LTCM. All of us are continuously predicting and forecasting. We have no choice. The future is uncertain. We plan; therefore, we predict and forecast. AI can assist in predicting and forecasting, to be sure. I would be disingenuous if I were to short shrift its utility. AI can lead to better decisions by humans. AI cannot, however, succeed independently of human oversight. Accuracy always distills to a person or people, not to a code and computer.
The qualitative aspects of predicting and forecasting are the reason it can't be otherwise. Human decisions are driven by intuition. Intuition is perfected by intelligence imbued with education, observation, curiosity, alertness, industriousness, relationships, and experience. The interplay between these variables is impossible to capture in an algorithm because it is continuously changing. Not even you can predict with fail-proof accuracy your actions tomorrow.
Time is another variable, and one so often of the essence. We must act immediately to exploit the opportunity our intuition recognizes. No computer can possibly be programmed to capture the indispensable uniqueness of human intuition and agility imbued in a profitable decision.
AI is neither artificial nor intelligence, but it is quite useful in situations were the variables the code incorporates are inert, constant, or predictable.
Your fingerprint and face, though not inert, are constant. They are fodder for AI analysis. AI can capture your fingerprints and face and decide with reliable accurately to let you in or keep you out. Traffic patterns are predictable: Morning and late-afternoon traffic is predictably heaviest. AI can direct you (as we see in the Waze app) to alternative, less congested routes These are but a sliver of examples of AI's utility.
AI is considerably less utilitarian when the algorithm is coded to act in response to human action. We humans are sentient, self-direct beings. Our tastes and preferences are protean. We interact with one another in ways that are unpredictable. That said, we have established habits, our daily life can be a repetitive cycle. Beware of geting too comfortable, though. The habits and cycles can change instanter and permanently on a whim, an impulse, a coercion, a rumor, a fact, a feeling (and they do).
Because markets are outgrowths of human action, they are assured to be intractably vexing to AI-directed computers. Just when you think you have it figured out -- the right habits and established pattern variables are captured -- it all changes.
Perhaps you remember Long-Term Capital Management, a 1990s hedge fund run led by Wall Street's best and brightest: Nobel-winning economists and ultra-alpha traders. LTCM's founders were convinced they had figured it out. They went to work to exploit their genius.
Using quantitative models, the fund employed an arbitrage strategy to exploit deviations in value between liquid securities across different markets. The strategy was successful out of the gate: The fund returned (after fees) 21% in its first year, 43% in its second year, and 41% in its third year. The computers were successfully generating profits based on the empirically observed repeating patterns of human behavior captured and coded in the algorithms.
Success out of the gate, but a failure down the homestretch. The geniuses overlooked the fact that humans are never predictable in perpetuity. Patterns change, and often in the time required to bat an eye. Algorithms, on the other hand, don't. The financial markets were hit with the Asian financial crisis in 1997 and then by the Russian debt crisis in 1998. LTCM's algorithms failed to accurately capture the immediate human reaction to both events. LTCM lost $4.6 billion in less than four months in 1998. The whole shebang collapsed soon after.
Fast forward to the here and now and we find a lesser AI failure in the financial markets.
Zillow announced this week it will shutter its high-tech, house-flipping segment. The segment relied on algorithms coded to forecast house prices. The algorithms would formulate offers to buy houses today based on the prices it forecast to sell tomorrow. Zillow's algorithms were apparently predicated on human behavior that would adhere to the established trend in home prices, which was up. Bloomberg reported that Zillow even tweaked its algorithms to generate buy offers at higher prices at the moment so that it could sell houses at even higher prices in the future. Such was the established trend, such was the coding embedded in the algorithms.
To forecast that human behavior will change unpredictably and inconveniently is the consistently accurate forecast. It is also the forecast Zillow's algorithms ignored, because the magnitude of change (implied in market volatility) is impossible to codify (accurately).
Homebuyer behavior turned more discriminating and price sensitive. Homebuyers are no longer willing to unquestioningly accept a seller's price as the trend predicted. Zillow now finds itself holding a multi-billion-dollar residential portfolio unlikely to fetch the algorithms forecasted sales price. Zillow acknowledged the new reality (and the algorithms failure) with a $304 million write-down in the latest quarter.
Were ignorance and parsimony to blame?
No. As with LTCM, Zillow was privy to a wealth of data, an abundance of intelligence, and millions of dollars of cheap capital. The favorable ternary was still insufficient to guarantee success. “We’ve determined the unpredictability in forecasting home prices far exceeds what we anticipated and continuing to scale Zillow Offers would result in too much earnings and balance-sheet volatility,” conceded CEO Rich Barton.
We are no different than Zillow or LTCM. All of us are continuously predicting and forecasting. We have no choice. The future is uncertain. We plan; therefore, we predict and forecast. AI can assist in predicting and forecasting, to be sure. I would be disingenuous if I were to short shrift its utility. AI can lead to better decisions by humans. AI cannot, however, succeed independently of human oversight. Accuracy always distills to a person or people, not to a code and computer.
The qualitative aspects of predicting and forecasting are the reason it can't be otherwise. Human decisions are driven by intuition. Intuition is perfected by intelligence imbued with education, observation, curiosity, alertness, industriousness, relationships, and experience. The interplay between these variables is impossible to capture in an algorithm because it is continuously changing. Not even you can predict with fail-proof accuracy your actions tomorrow.
Time is another variable, and one so often of the essence. We must act immediately to exploit the opportunity our intuition recognizes. No computer can possibly be programmed to capture the indispensable uniqueness of human intuition and agility imbued in a profitable decision.
RSS Feed