There’s a reason for that: The availability of and access to more computing power and data sets have ushered in breakthroughs of all kinds in everyday life, from cashierless grocery stores to voice-activated devices that respond to our commands from across the room.
The palpable excitement around AI centers on its potential to revolutionize seemingly every facet of every industry.
But AI isn’t a cure-all. Just because machines will eventually be able to learn almost anything asked of them doesn’t mean they should. So, what are the specific problems teams working on AI projects should focus on solving?
Here are three criteria my team follows.
Use AI to solve problems you can prototype first.
I suggest prototyping every AI application before commercializing it.
That will provide an opportunity to test, iterate and fail fast at a low cost and in a safe environment.
Without prototyping first, the product has a more limited ability to make a meaningful impact — and can even jeopardize your reputation.
While Waymo’s self-driving technology is now deployed in Chrysler minivans, its original home was in Firefly, a two-seater prototype vehicle that could not exceed 25 mph.
Waymo, an Alphabet company, shared its intent for Firefly to be “a platform to experiment and learn, not for mass production.”
Prototyping allowed the company to work out various kinks in low-risk settings like freeways before progressing to more complex situations like city streets.
Following the same approach in your AI initiatives can ensure your products have an established track record and a sufficient level of polish before they enter prime time.
Use AI to solve problems in applications where you can afford to make mistakes.
In order to continuously improve, we must create AI with a feedback loop that highlights when it makes the wrong decisions. Applying previous knowledge will continue to ensure smarter, more accurate assumptions.
This means you should start deploying AI in areas where the cost of making mistakes will not make a significant negative impact on your customer experience or reputation.
At our company, we first started using AI to streamline the home experience.
We use data from sensors and devices to understand occupancy in the home, then feed this information to our prediction engine to transition to away mode when the home is vacant, which turns off the lights as well as heating or cooling.
If the home makes an incorrect assumption and someone is actually home, the cost — in this case, a dark house and temporary thermal discomfort — is not severe and the solution — an adjustment in temperature settings and turning the lights back on — is simple.
As your team becomes more sophisticated at deploying AI, you can expand your use cases to scenarios with greater risk.
Use AI to solve specific problems, not entire systems.
People working on AI initiatives today generally want to make valuable contributions to society and as big of an impact as possible.
That’s why using AI to tackle many of the world’s deep-seated problems is top of mind: for example, personal transportation, health care and energy conservation.
Fortunately, intelligence does not have to be solved at a system level, as tackling specific problems is often more efficient and productive in the long run. Breaking the effort into smaller, yet significant projects also gives teams the ability to better allocate their often-limited time and resources.
Volvo showcases the benefits of this incremental approach.
The automaker unveiled autonomous emergency braking (AEB) several years ago to help prevent rear-end collisions and has since standardized this technology across the majority of its fleet.
While AEB will be a critical component of its autonomous vehicles, Volvo did not wait until it had a self-driving car on the road to introduce it. That decision helped advance its reputation as a leader in car safety innovation, driving stronger demand for its vehicles.
For all the promise that AI brings, we risk diluting its impact if we view it as a silver bullet.
Yes, machines are capable of accomplishing things that no human can, but not every scenario is the right application for this technology. We can only unleash AI’s full potential if we use it to solve the right set of problems.