What I Would Like AI to Do for Me

There is much concern lately that AI will displace human workers and perhaps eventually discard humans entirely.  I suppose such scenarios are imaginable.  However, I have been thinking about the potential of AI to do work that I currently do poorly.  I almost always under invest in this work, often resulting in poor performance, wasted time, and frustration.

AI should keep track of my things.  Keys, glasses, and phones are things I frequently misplace.  More ambitiously, I have 50,000 files on my computer in an immense hierarchy of folders.  When I am looking, for instance, for the draft essay I wrote some time in the early 1990s on art and technology, I would like AI to find it.  AI should be smart enough to know what to ask me as it tries to sleuth my files.

AI should know whom I know, as well as why I know them.  If something interesting, important, joyous, or sad happens to any of these people, it should tell me.  I know Facebook helps with this, but few of the folks I know use this.  LinkedIn might help AI with this, but I don’t want to have to log in and figure out the interface yet again.  AI should also read local newspapers where these folks live.

AI should understand my bank and retirement accounts, as well as investments.  It should provide totally tailored advice based on knowing my preferences, recent cash flows, and timetable of aspirations.  It should “keep score” on previous decisions, making deep inferences on what worked and did not work, and why.  It should learn from my reactions to these explanations.

AI should understand my health history and current state.  Beyond my health records, it should know my exercise records, purchases at grocery stores and restaurants, and be able to infer when I feel sick, tired, out of sorts, or depressed.  It should talk with me about my health and serve as a motivational health couch, with deep knowledge of what nudges motivated me in the past.

AI should understand economic and political trends and events, as well as evolving social forces.  It should know my preferences and inclinations, but not pander to them.  It should digest global news and address fake news like a shredder, sleuthing out culprits and motives.  AI could publish its own blog on its findings, but avoid divulging anything about me – see below.  It might develop its own persona in the process.

AI, as I have envisioned it, will obviously know almost everything about me.  As I age, it will likely know more about me than I do.  Why should I trust this broad and impressive set of capabilities?  What if it is hacked by adversaries?  With such questions in mind, I have drafted a very preliminary set of rules for cognitive assistants.

  1. AI must first and foremost serve my interests
  2. AI cannot unilaterally communicate with my network of relationships
  3. AI cannot tell anybody else anything about me
  4. AI cannot tell me anything that others have protected
  5. AI cannot reallocate or spend my resources without explicit permission
  6. AI, when hacked, must deny access and destroy risky content, with pre-planned backup
  7. AI, when hacked, has permission to move beyond defense to offense to destroy adversaries’ capabilities
  8. AI’s offense against a proven hacker is not subject to civil or criminal penalties
  9. AI cannot be subpoenaed to be a witness or testify against me in a civil or criminal trial
  10. Upon my death, my identity and AI’s knowledge of me will be managed in accordance with my Advance Directive

I am sure that I have missed things, or expressed them poorly.  Let me know your comments and suggestions.

Leave a Reply