Hacker News Reader

Comment by mjburgess | Hacker News Reader

Corporations have no interest in producing systems with the relevant properties to have "rights". AGI isnt of interest to business. Business is interested in automation; intelligence in this sense is precisely its opposite: a highly expensive, time-intensive, process of creatively responding to one's environment. It is, in a sense, the opposite of automation.

There's little danger a business is going to waste any time showing an AGI Paris over a long holiday, so that it can answer travel queries a little better.

Animal general intelligence is incredibly energy efficient, and it still takes a couple decades to produce anyhting "business useful". This is not a project basically any corp. is even working on.

the whole field of AI is, 99.99999% building a better calculator

View on HN

> Corporations have no interest in producing systems with the relevant properties to have "rights". AGI isnt of interest to business. Business is interested in automation; intelligence in this sense is precisely its opposite: a highly expensive, time-intensive, process of creatively responding to one's environment. It is, in a sense, the opposite of automation.

For a time, this will be true. But at some point a company with an AGI CEO will start winning, and then the tables will turn.

Enter a new AGI winter where economic incentives align against creating an obviously sentient AI because the social/legal implications prevent it.

We have 7bn incredibly energy efficient AGIs walking around the planet. It isnt clear to me we need any more; or have any incentives to create any at all.

Out of those 7bn, a few with capabilities significantly outside the norm can make a huge impact especially in inherently competitive scenarios. If you could use an AGI, even incredibly energy inefficient, with few other disadvantages and even very significant advantage over all of us 7bn, it can be tremendously valuable.

AGI means being responsive to an environment as we are. We already know the brain uses something like <<100W to do this.

There's a mistake implicit in all this. That technology is better than human beings. This is only true at everything other than intelligence. Machine systems are simply vastly more automatiable, and that provides the impression of their supriority. A train is "better" than a horse. But a train doesnt know what to do if it's lost.

Once you're actually looking at intelligence, animals are vastly superior to any existing technology in pretty much every respect. By orders of magnitude.

The idea that we could build an AGI to rival a fly is "years away". What you can accomplish with that teenager down the street, is vastly suprior to anything we can build.

It's a pointless project. Your base-case process is pregnancy. And technology is centuries away from anything close.

The human brain has many limits. It is limited in volume by the needs of birth, and it is limited in power consumption.

An AGI doesn't have those limits, so it could plausibly be better in some ways to the human brain, even if it uses up orders of magnitude more energy and volume.

The scaling limits are fundamentally different. That's why it can be worthwile.

Come on, the bandwitdh at which we can consciously consume complicated information and especially language is absolutely pathetic. Case in point: it takes the better part of two decades to educate a human to the degree where they are more or less useful to society. What is the most time consuming part of that process? Reading. One. Sentence. At. A. Time. Then spending more time integrating that information. An AGI has a huge advantage here because it is not limited to the senses it was born with. For example, it is significantly cheaper to train an ML model to detect credit card fraud than it is to educate and employ an equivalent number of people who can handle the incoming stream of transactions with comparable accuracy. It is not exactly automation, it is more like decision making.

AGI will happen eventually, but you're right. Even matching a fly is years away.

At best, we can approximate certain parts of it, ie, GPT3.

A quick Google says the human brain has about 86 billion neurons, each with up to 7,000 connections, with an adult having between 100 trillion and 500 trillion synapses [0]. Meanwhile, the commercial CPU with the most transistors is an Apple Duel-Die Ultra M1 with about 114 billion transistors [1]. If you equate 1 synapse with 1 transistor (Probably not a great metaphor), you'd need 1,000 CPUs to match a human brain, and the CPU could win out because of how quickly a CPU can "tick". Neurons typically only fire up to 200 times per second, whereas a CPU could do billions if the heat can be expelled.

[0] https://en.wikipedia.org/wiki/Neuron#Connectivity

[1] https://en.wikipedia.org/wiki/Transistor_count

Some government will eventually do it then when the tech seems within reach.

Reply on news.ycombinator.com