VGEL.ME

#InvisibleNetworks 09: memory emulator

Posted
a short story in the form of a screenshot of an email newsletter, from the blog "Bits and Bytes toward Eudomonia". full transcript: Responding to Dr. Zneser's Attacks. Paul Blackwell, Mar 7, 2032. Dear friends and subscribers, Most of you have probably now seen Dr. Zneser's op-ed on the research I published about in the last newsletter. I feel obliged to respond to the substance of his argument, despite its tone, as I deeply respect his prior work. In the introduction, setting out the core of his argument, Zneser writes: [quoted:] As humans, we are simply not prepared for this technology—it leads to increased misery, not happiness, not healing [...] [end quote] Ultimately, this is simply not supported by the research. He may go on to cite Plato and Maimonides, but he does not cite the statistics, because the numbers don't agree with him. 74% of users report "significant" or "very significant" reductions in feelings of grief after one session (Lee et. al, 2027). And the vast majority of users are able to accept that it is not real (Satchler, Green, 2028)—it is simply a tool for expressing feelings and working through what has happened in a supportive online community. I understand his consternation, but the world has changed a lot since those philosophers were alive. We live a lot longer than they did. We talk, chat, and video call over the internet. The average person interacts with AI daily, often more than with other people. We are a different culture than they were, one that is more used to and more capable of handling this technology. [quoted:] [...] consider the privacy concerns. Often, this process is initiated by a single person, yet the enormous quantities of data collected touch every facet of the target's life, and every relationship they had. All this data is hoovered up and sent to nearly unaccountable third parties [...] [end quote] Yes, the status quo here is not ideal. Training a simulator model is data-hungry, and often chat logs with just a single client are not enough to fully replicate a personality, requiring the acquisition of additional data. But the technology is advancing rapidly, and soon it should be possible to train these models on the client's device, obviating many of the privacy concerns of shipping highly-personal training data. [quoted:] [...] not helping. I missed him, not the model-facsimile of him. I missed our unrecorded small conversations in the middle of the night, our hugs—not our emails and pocket calls and text messages about where to get lunch. Yet I couldn't tear myself away—each DM, each tag, each video call from "him" picked at the open wound of grief, a wound that I couldn't allow to close over and scar because it would mean admitting that I had really lost him [...] I deeply regret my contributions to this technology, and I call on Dignitai—or the FDA—to halt the pilot program immediately. [end quote] I understand that Dr. Zneser had a difficult experience here—the Satchler study I mentioned above found that a small number of people (<4%) have trouble distinguishing between simulator models and the simulator target. We discussed this extensively last week in the comments, and the general consensus, that I share, is that we shouldn't ban an effective medical treatment simply because a small portion of the population has an adverse reaction. The mood stabilizer Lamotrigine is associated with a low rate of potentially-fatal skin reactions—if people experience them, we prescribe other mood stabilizers instead. Likewise, if someone is unable to distinguish a simulator model of their passed loved one from the real person, we should shut the model down and try alternative treatments. But we shouldn't take our eyes off the data: 74% report a reduction in grief! That is world-changing stuff! Real results! As I've said many times, I deeply respect Dr. Zneser. His research made training simulator models possible, and he was an early champion when much of the mental healthcare industry was still skeptical. But just because he developed this technology doesn't mean that he's an expert on its deployment and use. Just because he had an adverse experience doesn't erase the 74% of people who experienced a significant reduction in grief. In today's world, that kind of result is more important than ever. We need to keep our eyes on the prize. As always, towards a more hopeful and data-driven world, Paul
Twitter/X