Can AI Fix Anything?

Most people think AI is about replacing tasks. Reality is proving different. My experience applying AI, exposes how little we understand the systems we work in. Whether it’s students falling behind, military contracts going to the wrong bidder, or enterprise software failing to deliver, the real challenge isn’t tech. It’s the messiness of human behaviour, institutional inertia and a lack of trustworthy brands. This week, I reflect on what that means for AI in education, startups and the companies still stuck doing data entry by committee.

When to Use AI

The right time to apply technology is when it delivers an obvious outcome. The easier this is to demonstrate, the more likely adoption will be. Any objections should be non-technical.

Last week I touched upon the work we’re doing at MSBC in education and personalising learning for students. This fails this simple test for technology I just set myself.

Educational authorities in the UK claim that students perform as well as each other, once they’ve adjusted for socioeconomic background. In other words, as long as some children perform at the standards required, education gets a free pass because differences in achievement are caused by factors beyond teaching and the curriculum. It is this one-size-fits-all approach where AI offers transformational potential.

While it’s true that AI cannot correct the reasons for a poor start in life, if it helps to close the gap then I believe it’s worth pursuing. The challenge is that with many factors determining academic success, such as nutrition, a stable home life and emotional support, it is easy for opponents to find examples of where AI instruction falls short.

This is one reason why technology companies have made little headway automating teaching. Other areas offer more opportunities to win. One example is tunnelling, typified by Musk’s Boring company. The economic benefit of moving most traffic underground is considerable and the obstacles are human. The technology to bore tunnels is uncontroversial, while the reason it costs 10x more to dig in Manhattan than Paris is in large part political.

When Palantir got started 20 years ago, it spotted a way to consolidate military data that circumvented the billion dollar consultancy boondoggle in control of the process. It needed six months. A decade and a court case later, the company won the contract.

Around the same time, I had a similar experience on a much smaller scale. I was invested in a sportswear company that designed a shoe for the British Army. Officers were unhappy with the existing supplier and helped us craft a shoe to withstand the rigours of military training. Nonetheless, the incumbent won the bid with a shoe that failed the specifications of the tender. Our company lacked Palantir’s financial muscle to fight the state.

Creating efficiencies in the workplace requires a settled technology and full knowledge of human processes. Companies such as Salesforce have been collecting data on user behaviour for almost three decades. If that was all that mattered they should have won the AI race already. The fact that they haven’t is not because AI is unable to replicate and improve patterns of human interaction. It is because the data is not telling the full story of human activity.

Much like the example of education, where multiple factors contribute to a child’s performance, people produce different outcomes doing the same work. Repetitive data entry is easy to automate, but this definition covers only 60,000 of the UK’s over 30 million jobs. Data entry is more often part of a process, the outcome of which is determined by experience and intuition.

Startups find it much easier to introduce new technologies because they do not have existing ways of working. Established companies move slowly because of the risks of breaking what works. They have the advantage of brand, reputation and customers to buy them time to innovate. If you sell to them, you must earn their trust as well as prove your system works.

Two Takeaways

There are two lessons to emphasise here. The first is to collect data that tells the whole story about how people behave. This is straightforward on an e-commerce site where you know the offers that trigger people to buy. It is harder when researching a new opportunity, when internet search must be combined with conversations with existing players. In many situations and for the foreseeable future, AI is going to augment human activity rather than replace it.

The second lesson is the importance of brand. Technology that solves a problem is important, but it is nothing if people do not know who you are. Our sportswear company was an unknown quantity, even if our shoes performed best in testing. As more AI automations are rushed out as black box solutions for efficiency and scale, the trust factor behind any deal becomes more important. Brand reputation will remain as critical as ever.

Questions to Ask and Answer

1. Is the data I collect telling me how to improve my business processes?

2. What are the human factors that determine success in my business?

3. How does my brand create trust among buyers and suppliers?

 Here are 3 ways I can help:

1. Book a consultation to talk about AI.

2. A Guide to Automation Using AI.

3. Explore use cases using accelerated computing.