The US has the President it deserves
From Dave Winer:
In The Atlantic Tom Nichols writes that Trump is not a manly president. I don’t particularly care for that approach, I think honor and modesty are traits that should apply regardless of gender. We have the president we deserve. We’re the country that went to war without a draft, whose citizens got tax cuts while at war, whose citizens expect more of that, to us it’s never enough. We expect to be able to inflict chaos around the world and somehow never to be touched by it ourselves. That’s why people are out partying with abandon this weekend. They can’t imagine they can pay a price. There’s a reason Vietnam is responding to the virus so incredibly well and we’re responding so poorly. They remember fighting for their independence. To us, independence is a birth right. A distant memory that’s become perverted. We have to fight for it again. The virus is giving us that chance. We can’t get out of the pandemic until we grow up as individuals and collectively. Trump is the right president for who we are. We won’t get a better one until we deserve a better one.
What you need to know about Dominic Cummings
I’ve been reading Cummings’s blog since long before anyone had ever heard of him. Here’s what I’ve concluded from it…
He’s a compulsive autodidact. Nothing wrong with that, but…
He has sociopathic tendencies. (Some people who have worked with him might phrase it more strongly.)
His great hero is Otto von Bismarck. Note that, and ponder.
What turns him on are huge, bold projects carried out by people with vision, power, unlimited amounts of public money — and no political interference. Think Manhattan Project, Apollo Mission.
So basically he’s a technocrat on steroids.
He regards most professional politicians as imbeciles.
Like many fanatics, he has a laser-like ability to clarify, and focus on, objectives.
Johnson can’t get rid of him, because without Cummings he hasn’t a clue what to do. And that’s a big problem in the longer term because…
Cummings knows how to campaign, and how to plan projects where there is hierarchy, authority and autocratic control but…
He knows nothing about how to govern.
And neither does Johnson
This will not end well.
Stuart Russell’s Turing Lecture
The Annual Turing Lecture was reconfigured to take account of Covid and so today was delivered remotely by Stuart Russell from California. The whole thing was recorded — link here. In essence it was a brief recap and extension of the arguments in his book Human Compatible: AI and the Problem of Control.
Russell is a really major figure in the field of Artificial Intelligence (his and Peter’s Norvig’s Artificial Intelligence: A Modern Approach is still the leading textbook), so the fact that he has become a vocal critic of the path the discipline is taking is significant.
Basically, he thinks that our current approach to the design of AI systems is misguided — and potentially catastrophic if in the end it does succeed in producing super intelligent machines. That’s because it’s based on a concept of intelligence that is flawed. It assumes that intelligence consists of the ability of a machine to achieve whatever objectives it’s been set by its designers. A superintelligent machine will achieve its objectives without any concern for the collateral damage that what might wreak. The elimination or sidelining of humanity might be one kind of collateral damage.
Russell uses a nice contemporary example to illustrate the point — the recommendation algorithm that YouTube uses to compile a list of videos you might be interested in seeing after you’ve finished the one you’re watching. The objective set by YouTube for the machine-learning algorithm is to maximise the time the user spends watching videos by finding ones similar to the current one. And it’s very good at doing that, which has some unanticipated consequences — including sometimes luring users down a wormhole of increasingly extreme content. The fact that YouTube has had this property was not the intention of Google — YouTube’s owner. It’s a consequence of the machine-learning algorithm’s success at achieving its objective.
The problem is, as Russell puts it, that humans are not great at specifying objectives wisely. It’s essentially the King Midas problem: he wanted everything he touched to turn to gold. And his magical ‘machine’ achieved that objective. Which meant that in the end he starved to death. And the smarter the AI the worse the outcome will be if the objective it is set is wrong.
If AI is not to become an existential threat to humanity, Russell argues, then it has to take the form of machines which can cope with the fact that human purposes are often vague, contradictory and ill-thought-out, and so essentially what we need are machines that can infer human preferences from interaction and human behaviour.
It’s an intriguing, sweeping argument by a very thoughtful researcher. His book is great (I reviewed it a while back) and the lecture introduced it well and set it in a wider and comprehensible context.
It’s long — over an hour. But worth it.
Quarantine diary — Day 66
This blog is also available as a daily email. If you think this might suit you better, why not subscribe? One email a day, delivered to your inbox at 7am UK time. It’s free, and there’s a one-click unsubscribe if your decide that your inbox is full enough already!