Are you ready for HAL?: 4 questions to ask about AI before launch

3 years ago 424

Although the fictional HAL supercomputer was archetypal introduced to movie-goers much than 50 years ago, determination are important lessons learned that AI practitioners tin use today.

HAL (heuristically programmed algorithmic computer) archetypal debuted successful the Stanley Kubrick classical movie "2001: Space Odyssey" (1968). While portion of HAL's programming required the machine to support the existent intent of the ngo a concealed from astronauts, HAL was besides programmed to assistance its quality travelers connected the ngo by verbally taking questions and instructions and besides providing verbal feedback with the assistance of natural connection processing.

SEE: Hiring Kit: Video Game Programmer (TechRepublic Premium)

During the voyage, HAL experienced logic conflicts erstwhile it attempted to equilibrium relaying captious accusation to astronauts against its directive to support ngo accusation secret. The extremity effect was a bid of bundle malfunctions that placed HAL connected the way of destroying the quality inhabitants of the vessel successful bid to safeguard the secrecy of the mission.

"2001: A Space Odyssey" showed successful theaters much than 50 years ago, but is prescient successful the questions that loom for organizations arsenic they inject artificial intelligence into concern processes and decisioning. Among these questions are:

What's accurate?

In October 2019, Amazon's Rekognition AI mistakenly classified 27 nonrecreational athletes arsenic criminals, and successful March 2021, a Dutch tribunal ordered Uber to reinstate and compensate six erstwhile drivers who were fired based connected incorrect assessments of fraudulent enactment that were made by an algorithm.

Many organizations participate the AI arena by purchasing an AI bundle that is already pre-programmed by a vendor that knows their industry. But however good does the vendor bundle recognize the particulars of a circumstantial firm environment? And if companies proceed to bid and refine their AI engines, oregon they make caller AI algorithms, however bash they cognize erstwhile they're inadvertently introducing logic oregon information that volition output flawed results?

SEE: Gartner: AI is moving accelerated and volition beryllium acceptable for premier clip sooner than you think (TechRepublic)

The reply is, they don't cognize due to the fact that companies can't observe flaws successful information oregon logic until they observe them. They admit the flaws due to the fact that of their empirical acquisition with the taxable substance that the AI is analyzing. This empirical cognition comes from on-staff quality taxable substance experts.

The bottommost enactment is that companies indispensable support quality SMEs astatine the extremity of AI analytic cycles to guarantee that AI conclusions are reasonable—or to measurement successful if they are not.

What's ethical?

A ample retailer wants a predictive bundle that tin expect lawsuit purchasing needs earlier customers really marque purchases. The retailer purchases and aggregates lawsuit information from a assortment of third-party sources. But should the retailer acquisition healthcare accusation astir consumers to find if they request diabetic absorption aids?

This is an morals question due to the fact that it intersects with idiosyncratic healthcare privateness rights. Businesses indispensable determine the close happening to do.

Where bash humans acceptable in?

In the end, quality cognition is the operator of what AI and analytics tin do.

The modular is that AI is cutover to accumulation erstwhile it is wrong 95% accuracy of what taxable substance experts would conclude. Over time, it is apt that this synchronization betwixt what a instrumentality and what a quality would reason volition drift.

SEE: Deloitte: The apical concern usage cases for AI successful 6 user industries (TechRepublic)

Realizing that AI (like the quality brain) isn't ever perfect, astir organizations opt to person a taxable substance adept arsenic the last reappraisal constituent for immoderate AI decision-making process.

What limitations bash we face?

Today's AI analyzes immense troves of information for patterns and answers, but it doesn't person the quality quality to intuit oregon tangentially get astatine answers that aren't instantly successful the data. Over time, determination volition beryllium enactment to heighten AI's intuitive reasoning, but the hazard is that the AI tin spell disconnected the rails similar HAL.

How bash we harness the powerfulness of AI truthful it does what we inquire it to do, but doesn't extremity up blowing the mission? This is the balancing constituent that organizations utilizing AI person to find for themselves.

Data, Analytics and AI Newsletter

Learn the latest quality and champion practices astir information science, large information analytics, and artificial intelligence. Delivered Mondays