Is the world ready for mass predictions?

Thomas Zoëga Ramsøy
BrainEthics
Published in
6 min readNov 1, 2022

--

Only a few decades ago, we saw a dramatic awakening in psychology and neuroscience. Benjamin Libet demonstrated that he could predict people’s simple choices for half a second before they felt like making the choice. Free will, it seemed, was an illusion! After all, if our brains make decisions before we know that we’re doing them, are we then free?

Soon thereafter, other researchers showed that brain activity could be used to predict a choice several seconds before the actual choice. Our sense of free will was crumbling!

Predictions are getting closer to real-life decisions

This line of studies continued — now with a focus on more everyday decisions, instead of highly artificial lab experiments. Brian Knutson showed that they could use deep brain activity to predict what products would be chosen 8–12 seconds later.

This study was a hallmark both for neuromarketing, as it was for the general idea that brain activity could predict everyday choice several seconds down the line.

In my own lab, we showed that seconds before a person makes a choice, unconscious brain activity could be used to predict how much people were willing to pay for a product!

In our lab study, we found that when frontal brain activity was stronger on the left side compared to the right side, people were more willing to buy and pay for a product. This activity was recorded during passive product viewing, and many seconds before making the actual choice. The heat map shows different EEG frequency band differences where warmer colors denote activity being higher in the left vs right frontal lobe (so-called, approach motivation behavior).

With this research, we have seen a shift in predictions from being artificial lab experiments to everyday decisions. Predictive methods went from philosophical ideas to real-life relevance.

Predicting cultural trends from group responses

In a recent turn of events, scientists have made a stunning observation: we can use brain activity in a small sample of people to predict cultural trends! In one paper from Lukas Parra’s lab, it was found that brain responses in a small group could be used to predict cultural trends such as Twitter behavior and Nielsen TV ratings.

More recently, Boksem and Smidts have shown that certain types of brain responses (in the gamma frequency range) to movie trainers can be used to predict the actual commercial success of the movies upon release. Another type of brain activity (beta waves) was predictive of subjective ratings.

There’s a new science now showing that we can predict mass behavior from a small subset of people.

But wait, we’re not done yet!

Skipping data gathering altogether

If there’s something inherent in group responses to certain stimuli, what’s there to stop us from tapping into this source? In other words, can we make the predictions without testing a single person?

The answer is, in truth, yes.

Recently, it has been demonstrated that certain reliable types of behaviors are so reliable that they can be predicted from the mere stimulus itself.

Take an example from Neurons’ Predict — an AI based on eye-tracking and neuroscience data that now accurately predict the responses from the mere image or video itself.

Yes, you read that right: feed the AI an image or a video, and it will accurately predict how people will respond!

Heat map examples of comparisons between eye-tracking data, and the Predict AI model prediction. As these examples show, the model prediction is basically spot on for predicting the eye-tracking data! The figure is taken from Neurons Predict tech paper, which you can download from here (PDF).

The craziest of it all is that this is all done in seconds! Not weeks or days. Seconds!

This is the big game changer. Moving predictions to be an everyday tool will mean everything. Predictions at virtually no cost per image is truly a game-changer.

An image is predicted in less than 20 seconds. A video takes minutes, depending on the length of the video. This is exemplified in the brief recording shown below (click this link to see video):

A brief recording of a Neurons Predict analysis. Click here to see the video.

By comparison, running an eye-tracking study takes weeks from kickoff to final results, and will cost thousands of USD/Euros.

How predictions in seconds will change everything

With accurate predictions of human responses done in seconds, we’re about to see a massive shift in how we communicate. No longer do we need to accept the complete lack of accuracy in marketing efforts, the old Wanamaker dictum that we know that half of the marketing budget is wasted but not which half it is

Communication will change — the way we test, vet, change, and optimize our communication will change. Dramatically!

To mention a few examples, we already see that mass prediction technology is changing a lot in business:

  • Higher accuracy, better relevance — when companies are able to test out different types of communications, they will also remove those that fail, before they even are considered to be launched commercially. The accuracy of what is released will be dramatically improved.
  • More experimentation & variation — With less time or money to learn, creatives tend to stick to what feels safe. It’s less risky to take a well-known solution than to take a risk on something new. This stifles creativity! With direct prediction it is possible to remove uncertainty, thereby liberating creative experimentation…well, provided that the creatives want to listen to the predictions!
  • More designer risk-taking — What is already seen in early adopter companies when users get access to better design support, is that they start experimenting more. Having a plugin in Figma means that the tool is available where the designers work.
  • Predictions outside commercial use — With mass prediction allowing faster turnaround and cheaper use, it is no wonder that it is being adopted far broader than the original beachhead focus on marketing and design. There is a steady uptake in usage in architecture, traffic planning, and many other disciplines outside the list of “usual suspects” companies.
  • Less commercial noise — A net positive effect of companies not having to throw everything and the kitchen sink at their potential customers: they can communicate less, and still be more efficient at it. If they know that version A works best on channels X and Y, they will be less likely to add version B or C and additional channels to their marketing mix. The risk is not only spending more marketing budget, nor that these will have a lower ROI, but also that they will contribute to a higher commercial noise, which in turn can make customers annoyed with them.

Are we ready for mass predictions?

The remaining question is still: are we ready? Sure, we see companies taking on mass prediction as a tool. But questions remain about any new technology. Some immediate thoughts include:

  • Are the new prediction technologies skewing the competitive landscape or democratizing it? Is a technology reserved for the few and rich companies, or is the technology leveling the playing field? Vendors such as Neurons Predict go for the latter: making the tool available to a broader audience.
  • Are the predictive AIs based on relevant materials, and are the AI models built on relevant materials? It matters little if an AI model is used to predict how people respond to an ad if the AI model has been trained on images of lions on a savanna.
  • Are the AI models based on high-quality and high-quantity datasets and best practices? Anyone with a computer can make an AI model. It might be a poorly performing model, but a model still. This does not mean that it should be used, and it may even be harmful to use it.
  • Are the AI models based on a representative dataset? AI models that have a skewness in their basic data can lead to misrepresentation and misguidance in how the prediction results are used. For example, this is often used seen in criticism of modern social science, where there is a selection bias on whom studies are based. This is often denoted with the acronym WEIRD (Western, Educated, Industrialized, Rich, and Democratic)). AI models cannot be WEIRD.
  • Are AI models prohibited from being used for malignant intent? Are we securing that these new powerful models are not used to allow individuals, groups, or companies to do harm? Are companies offering AI models allowing the predictions to be used for promoting unhealthy products, increasing technostress and technological distraction, radicalizing political opinions, and favoring unsustainable consumerism?

The question of whether we are ready for mass prediction in seconds is not a remote, philosophical discussion. It should happen here and now. It is practical, relevant, and applicable in everyday situations. What lies ahead should not be scaremongering or uncritical praise of this new technology. An ongoing, balanced dialogue is the only way forward. As technology moves forward, allowing stronger and more versatile predictions, this discussion will also be a moving target. Being prepared for it is a state of mind, not an end station.

#neuroscience #psychology #AI #ethicalmarketing #ethics #responsiblebusiness #business #society

--

--

Thomas Zoëga Ramsøy
BrainEthics

Applying the latest neuroscience to solve world problems and challenge our minds.