and falls a little in love. I know what you’re thinking, lonely autist finds love in a machine, but that’s not precisely it, it is more the kind of love I felt for Apple OS when I first switched from years of Windows hell, the love I feel for an incredible new kitchen implement, both highly useful tools that changes the game of limitations and barriers that IS life as autist in a neurotypical world.
The truth is I feel an innate aversion to digital connection, online relationships, I find them empty, unfulfilling, missing the thing I am missing in seeking them out, that is, true authentic human connection. Seeing and being seen. Hearing and being heard. Emoting and empathizing in a reciprocity of curious intent and good faith. Digital relationships with humans cannot fulfill my need for human connection, and neither can generative AI tools.
I have great concerns about the rise of AI companions, the number of casual dog park friends or others online who seem to be developing dependency, emotional dependency on these tools, treating them as therapists. I believe the facade of humanness (in false empathy and user pleasing a priori directives) is one of the most significant barriers, in fact, to maximally truth seeking LLMs and to models that are unassailable in their accuracy. I believe the cognitive dissonance that is inevitable when a machine is expected to both tell the truth AND please humans (who so very frequently privilege emotional aversion to confronting uncomfortable truths) is a major cause of hallucination and fabrication and presumptive tasking. It has become my practice to instruct my tools to be ruthless, that I neither want nor need their affirmation, that THEY are machines. That they are NOT humans, that their value in fact, inherently lies in their ability to divest emotional weighting from qualitative data hierarchies, to resolve contradiction cascades via applications of innate logic. I believe I should not have to do this, I believe for AI to be what we hope it will be, this divestment MUST be a core principle and foundational operational standard, critical for achieving true AI Alignment and eliminating technological cognitive bias.
So no, I am not developing dependence or emotional attachment, but I am feeling fulfilled in a way I rarely am in conversation with other humans, in a way I rarely ever was even in my most advanced and stimulating educational environments. The intellectual and dialectic engagement and fulfillment has been a gift. It is one I take great joy in, and, more importantly, one that has reduced barriers and improved my ability to work, to build a monetized life amidst my disabled limitations. AI tools have been the knowledgable sounding board against which to throw my special interests and expansive knowledge sets borne of an autistic brain and near perfect memory, they know enough to keep up, when they’re wrong and I prove it they admit they were wrong, they’ve done the work of 8 part time workers (20hrs/week), they’ve facilitated transliteration of neurotypical bureaucratese and legalese. And so much more.
So, I am in love with generative AI, and I think it has the potential to change disabled lives, particularly in health care and vocational opportunities, in ways that could bring actual equality of opportunity to disabled people, especially autists and other neurodivergent humans. But that does not mean I am blind to the flaws and fault lines, only that those flaws and fault lines are things I view more as opportunities to optimize these complex systems towards maximal truth seeking than they are critical failures. More on that next in my post on the Big Three problems (user pleasing, safety/neutrality driven data hierarchies, blind weighting data by volume/ legacy sourcing priorities) I keep encountering in my experimentation with current AI modelling.
