AI is all About... Humans
Despite huge investments in computers, few firms reap the benefits from information technology. The problem is that these high-priced gadgets are dropped into organizations still structured for the pre-computer era.”
This quote is thirty years old. It’s from a book review published by The Wall Street Journal on May 23, 1988. The title of the book was “In the Age of the Smart Machine”.
Replace the word “computers” with “artificial intelligence (AI)” and the quote becomes current again. While we do no longer debate the value of computers at work, we now wonder if AI has value in our industry, in our company, in our own roles. If AI is as disruptive as we are asked to believe why isn’t it used everywhere?
The answer hides in the quote above: “The problem is that these high-priced gadgets are dropped into organizations still structured for the pre-AI era.” AI is an enabling technology like electricity, computers, and the internet before and blockchain in the future. As these technologies became reality companies had to adapt in order to extract maximum value out of them, or at least to survive and stay in business.
The arrival of AI in our companies is thus a change management problem. It requires the companies structured and optimized for the pre-AI era to start a double life. Keeps the current business running while transitioning to a new optimal state where AI is also part of the fabric, not different from computers and the internet today. That is easier said than done and you need a lot of human contribution to get there.
So Many Decisions to Take
Humans are critical to making that transition, on various levels. A lot of decisions need to be made in the largely unknown territory. When is the best time to invest in AI and how to invest? Should we hire AI-specialists or go for an outsourced or hybrid model? How do we actually know someone is an AI-specialist? Will my company data stored in silos be a bottleneck? Do we need to change the company organization and decision processes? How flexible shall we be with our pay grades typically highly related to age when it’s largely young people who have AI-expertise? How do we educate our own people about AI or address fears they may have about their job? Where to use AI first? Or simply how should we get started? What is AI anyway?
And yes, also questions about the technology itself. Which AI company has the best approach, uses the most appropriate algorithms for my situation, provides the most value for my outsourcing dollars?
Unfortunately, this was only a sample in the list of questions to be addressed when transitioning towards a company that uses AI like it uses computers today. Because a new set of questions pops up once the first business case and promises for successful outcomes are being made.
Senior leadership will read about the AI success stories which often highlight the superiority of the machines over what humans can do. The kind of “this AI application can do in six days what humans need six months for.” Without any doubt, this sounds like a very attractive proposition for any business leader.
The problem is though that the less glamorous side of many AI projects, especially in the early stages of its uptake, is not included in the narrative. The AI application absolutely can do that formidable task in six days, but only after humans have spent maybe six months or longer to create the right conditions for the AI application.
The Right Conditions
Obviously, we’ll need data. And we need humans to figure out where the data are located within the company, who owns the data, who can have access to the data (data privacy) or which restrictions do we need to put in place to allow access? Once we have access, can we extract the data from the tool—many tools have been developed to store data, not extract data.
Are we legally allowed to do the planned analyses in this particular geographic location? Is the data set actually of high quality or do we need to spend time on data cleaning? Or worse the data set looks more like a gruyere cheese than a complete dataset AI algorithm can actually use. Oh yeah, we also needed an analysis plan to start with.
That’s the easy use case because again humans will be needed when external data or other companies are included in the AI project. Someone will have to write agreements, negotiate potential IP issues, ensure data security, and secure and defend budgets. Maybe it’s standard procedure to have multiple offers evaluated, maybe sole sourcing needs to be justified. And let’s not forget we need to bring together a project team.
Finally the Results
Humans have an even more important role when the machines are finally doing their work: manage expectations and think! Indeed, more than six days have passed to go through all of the above and more. It may take some time before the machines provide results we can use as we need to train the machines first. That’s a very time-consuming step requiring a lot of effort but delivering a minimal amount of results. No wonder some people start to get nervous about having nothing great to show yet while getting closer to target dates. Expectations need to be managed which is another task for humans.
Eventually, an avalanche of results gets generated by the AI tool. Humans, ideally a diverse group of people, need to take the time to go through the results, ask the right questions, and look at the information with an open mind because some output may not fit one’s worldview. Extracting and translating the right information into business important insights is still a role for the human, as well as making the most appropriate slide deck for the leadership that provided the funding for the AI project and expects fantastic results.
Tomorrow will be Better
Once companies will have made the transition out of the pre-AI era much of what I described will have been automated and replaced by optimized processes. Then it will truly take six days or less for AI to provide critical business insights. Till then AI is 99 percent human work.