What are the Biggest Companies in Tech Saying about Empathic AI?

Emojis Fausto Garcia 1920X1080

Although we work with companies at different stages on the path to building empathic technology, the ability for machines to truly understand their users and provide valuable empathic interaction is far from a solved problem. In our discussions with businesses in all our market sectors, there are a few issues we hear frequently repeated. Even the leading players are grappling with similar challenges to their trailing competitors. We recently went out to the market to discuss how best each company would want to integrate our developer tools into their product development process. We’re sharing the most salient insights from those conversations here to air them out in the open. Let’s all take a minute, hold hands and say, ‘it’s OK, you’re not alone’!

The hot topics that we found across every conversation were:

  1. Sensor selection: what sensors should we choose to get what kind of results?
  2. Sensor fusion: one data stream can be useful, many are insightful.
  3. Integration/procurement path: there isn’t one flavour, so you need to be flexible.
  4. Ethics, privacy and rights: how can a brand work in favour of customer & human rights?
  5. Hardware vs software: should empathic AI be bundled in physical machines, or as code at the edge or in the cloud?

Microphone Daniel Sandvik 1920X1080

1. Sensor Selection

Choosing Sensors that provide the right mix of data for robust measurement is problematic. Before committing budget, teams want help navigating the compromises of configuring their sensor suite. There are pros and cons at all levels; across the types of sensors and data streams they produce, sensor quality, efficiency, as well as usability & privacy concerns.

There is also future-proofing to consider. A market-leading sensor today might be usurped by a higher quality competitor tomorrow, or a new method of measurement might come in to replace it. For example, heart rate is usually best measured by a sensor in contact with the skin, preferably against the chest, next to the beating heart itself. But that can be impractical or invasive for users. A wrist-worn device might be an acceptable alternative but the signal quality is likely to be lower. After that, contactless options such as infrared video bring the advantage of minimising intrusiveness but their quality may be lower still. Then, entering the market from left-field right now are alternatives such as biometric radar sensors. The point is, you might place your bets on one option only to find another available shortly after.

You can read more on sensor selection in our recent article, Choose Your Sensors for Empathic AI.

2. Sensor Fusion

The most useful part of the empathic technology stack for businesses currently developing their own seems to be sensor fusion. This is perhaps just for the simple reason that it’s hard. You have a collection of sensors, along with other data sources, all running on different protocols, frequencies, and so on, and you need to fuse them together in a way that allows you to run both individual and combined analysis on them.

This is something we’ve been working on since our inception, broadly taking an approach that mimics human cognition. Our bodies have sensors that receive information from the outside world in various formats, and we have to choose which ones to pay attention to from one moment to the next. This ‘multimodal’ way of processing fuzzy information from multiple sources has been our modus operandi for measuring human states since the start.

Here’s a bit more background – Sensor Fusion: The Only Way to Measure True Emotion.

3. Integration/Procurement Path

There seems to be a broad spread in how companies expect to acquire empathic AI solutions. Some are looking to build their own, others to co-develop it in partnership with a specialist supplier, and others still who are happy just to purchase a solution either directly from a producer or through an established industry supplier who is licensing it.

4. Privacy and Rights

Brands are made up of the humans that work for them. So the concern over how a person’s personal data is recorded, analysed and reacted to is a human one. No humans, or brands that they work for, want to be in a position of breaching human rights for the sake of a commercial return. Or at least none that we’ve talked to, and these are some pretty large brands. Having built and managed complex technology for decades, they seem to be treating the ethical challenges with the rigour they’re accustomed to – slowly, deeply and cautiously.

While this is a complex issue, we see a simple framework underlying it, harking back to the basics of universal human rights. Don’t take control of someone’s freedom to choose. Build relationships. Don’t be data authoritarians. If brands act as good data citizens in the first place, they are left with a more manageable challenge that is essentially one of UX.

For some broad thoughts on this, read Digital Humans Need Digital Ethics.

5. Hardware vs Software

We are seeing variation across the formats in which companies want to onboard empathic AI for integration into their own products. At one end of the spectrum, some teams want us to send them a mini-computer with all the software preloaded, with or without a set of our own recommended human-data sensors, along with our online analysis dashboard for data review. At the other extreme, some teams just want libraries they can load into their own systems to do it all themselves.

The other axis of this decision space addresses the need for local or cloud-based services. Again, there is broad variation. Some teams are happy communicating with our cloud API for online processing, accepting potential compromises of lag and connectivity in return for a simple and remotely-updatable solution, while others want to run the processing entirely locally, on the edge, inside their product (eg. car, laptop, etc.). These parameters of operational tolerance generally correlate with the development phase the customer is at, from exploratory research, through gathering data for modelling, to preparation for deployment.

We’re very keen to learn more about the market’s varying needs and objectives in this space, as it feeds into the grand challenge of designing smart IoT infrastructure that operates optimally both on- and offline while maintaining data privacy, tolerable connection speeds and so on.

So where are you on the journey to making your products empathic, and how will you transition to the next step? We’d love to talk about that.


===

Content credits:

Ben Bland

Chief Operations Officer