Making technology and innovation predictions

ICON_PLACEHOLDEREstimated reading time: 7 minutes

In the news, on social media such as LinkedIn and other places, I notice a lot of technology and innovation predictions about what will certainly happen within the next ten years. AI here, nano-something there…

Making predictions is hard. Very hard. Particularly if one tries to predict things ten years in advance. Good innovation analysts would probably shy away from such long-range predictions.

But there seems to be some hope of being able to make predictions up to, say, one year in advance. For example, Philip Tetlock and Dan Gardner describe very interesting cases and methods in their book Superforecasting. Many of the findings in the book came out of The Good Judgment Project. The Good Judgment Project was a very successful participant in the IARPA-funded ACE Program. This program aimed at investigating and improving the accuracy of intelligence forecasts.

So, OK, one year. But ten years…

One-month vs. ten-year predictions

If we accept that making predictions ten years ahead is a lot harder than making predictions one year ahead, we should see fewer ten-year predictions than one-year predictions. We may even introduce more intermediate levels:

• one month ahead
• six months
• one year
• ten years

Since predictions get harder the further ahead one aims to predict, I would expect to see most predictions for one month, and the fewest for ten years ahead.

I used Mergeflow’s Delta-t tool to test my hypotheses. Delta-t lets you explore how topics develop and emerge over time, across various signals (R&D, technology blogs, VC investments, patents, industry news). Given Mergeflow’s search space, which focuses on technology and innovation contents, the results will be from this domain, and not, for example, predictions about who will win the next Academy Award or so.

For my searches, I used combinations of search strings like “within the next decade”, “within the next ten years”, etc. to capture the ten-year predictions, and analogous search string combinations for the other prediction time spans.

Below are the results, in a screenshot from Mergeflow (click on the image for a larger version). The screenshot shows how the shares of the different prediction time spans developed over the past four years, across R&D, technology blogs, VC investments, and industry news (I excluded patents here because they do not really talk about predictions):

Whoa! All the way from 2014 to 2017, the results were the opposite of what I expected. Instead, the share of one-month predictions is the smallest, and the one-year and ten-year predictions have the biggest shares!

Technology and innovation predictions ten years ahead

Who makes predictions ten years ahead?

In Mergeflow’s Delta-t tool, you can zoom in on all the underlying data for each topic. So first I looked at where the ten-year predictions come from:

Ten-year predictions are mostly to be found in Scientific Publications and Technology Blogs, as it turns out. Let’s look at these in more detail. Starting with Scientific Publications, what are the most prominent technologies that feature in ten-year predictions? Here is what Mergeflow extracted from the data:

Some of the usual suspects, it seems. For example, the topic complex of machine learning, big data, data mining, and neural networks, where one of the predictions being discussed is the “possibility of achieving, within the next decade(s), full simulation of the human brain” (go to the paper). And immunotherapy, where one of the predictions includes that “within the next decade, it is likely that epigenetic modifications may be used as biomarkers to aid in diagnosis and treatment of food-allergic patients” (go to the paper). Well, somebody could check in ten years from now to verify these predictions.

How about technology blogs, another important source of ten-year predictions? Here are the top technologies extracted by Mergeflow:

Of course, artificial intelligence. My personal favorite there is this headline: Experts Say AI Has a 50% Chance of Beating All Human Intelligence Within 45 Years. OK, so this even goes way beyond the next ten years. Although I think that “50% chance” is about as noncommittal as it can get.

Power plants are another big topic in blogs. The biggest subtopic there is nuclear power. For example, one article argues that a new type of reactors could “make fusion power practical within the next decade” (fusion, not fission). Haven’t we heard this before? Ten years ago, perhaps? Like here?

Technology and innovation predictions month ahead

Let’s get more modest now, and look at predictions one month ahead. Here is the distribution of one-month predictions:

As you can see, this is different from the ten-year predictions. The latest overall share is 9.5%, compared to 39% for ten-year predictions. But be careful when you visually compare the ten-year and the one-month chart. Their y axes scale differently because otherwise one would not see anything in the next-month chart. The distribution across Scientific Publications, Tech Blogs, VC Investments, and Industry News is different as well. Fewer science and fewer blogs. Zooming in on the blogs, for example, shows these technology terms in Mergeflow:

So it is mostly about smartphones and their cameras (cf. terms such as “lenses”, “display resolution”, and “aspect ratio”).

“Next month” in my Scientific Publications result set is mostly about quitting smoking. For example, there is a study that aims to predict post-stroke smoking behavior.

All in all, next-month predictions seem to be more about relatively down-to-earth things like product releases or behavior changes. The advantage of these things is that they are easy to test (either the product was released within a month or not; either one changed a behavior or not).

Lessons learned

So what did I learn? First, I really was surprised to see so many more far-ahead than around-the-corner prediction conversations. But then again… it is probably a lot easier to generate attention-grabbing headlines with things that are far out and hence sound more spectacular. To me, a question that remains is, should I predict that within the next ten years, people will keep making more ten-year predictions than one-month predictions? At least, this prediction would be easy to test.