Article (6.07) Leadership and Technology, for the Wider Human Experience

A man builds a device so the woman he loves, who is losing her sight, can write to him without dictating her words through another person’s hand. It becomes the typewriter. The typewriter becomes the keyboard. Almost nobody using one today has any connection to blindness.

Pete Horsley told us that story. He is the founder of Remarkable, the disability tech accelerator created by the Cerebral Palsy Alliance, and he tells that story as a working principle rather than a curious trivia factoid. Assistive technology, in his account, isn’t a side category built for a smaller audience and occasionally borrowed by everyone else. It’s often the first draft of technology everyone eventually uses. Closed captions were built for deaf and hard-of-hearing viewers; now they run on trains, in houses with sleeping babies, on phones with the sound off. A computer-vision game controller built by one founder so his brother, who has cerebral palsy, could play games with a head tilt or a poked-out tongue, now sits inside the accessibility settings at Salesforce and Microsoft.

Pete’s argument is that this pattern is not incidental. Constraint can force someone to solve a harder problem than the market was asking for, and the resulting product often works for people the market wasn’t thinking about either. He calls this designing for the “wider human experience” – not a euphemism for inclusion as a virtue, but a description of who a product actually needs to serve if it’s going to hold up. The alternative, he suggests, is quietly optimising for a narrow slice of the population and hoping nobody outside it ever needs anything else.

Pete is explicit that Remarkable isn’t a charity dressed up as a business: it takes equity, it backs founders, it wants an exit. The commercial case has to hold on its own, because goodwill alone won’t fund an accelerator through the years it takes a startup to find its market. That case gets harder to make with artificial intelligence, where the training data itself often reflects the same narrow slice this project is trying to design around. Facial recognition systems can fail to register facial difference. Pattern-matching systems built on data that has, in effect, expunged its outliers can carry that omission forward into whatever they generate next, especially now that some models are training on their own synthetic output rather than fresh human data.

There’s a genuine tension here, one the conversation doesn’t try to resolve too neatly. That is, lived experience is difficult to reduce to training data, though training data is often what determines whether a system works for the people whose experience never made it into the dataset. His answer rests less on philosophy than on demographics. By 2040, roughly a fifth of Australians will be over 65, and around half of that group will live with some form of mobility, cognitive, or hearing impairment. A model built without that population in mind isn’t a model with a gap. It’s a model with an expiry date.