These "agents" of which you speak
Do any of them resemble some kind of High Elf King?
Artificial intelligence agents can invent their own language and talk among themselves to work out the best way to get a job done, a study has shown. Conversations come natural to humans, but they are a massive challenge for computers. Recent successes in this area include better human-language translation, and simple question …
I am trying to remember if it was Clarke or Asimov who had a giant computer that was hidden underground in a secret location and needed other computers and technicians to translate between it and people.
Of course there is also the scene from Person of Interest were Finch reveals it took 43 goes to create an AI that didn't try to escape and/or kill him.
"I am trying to remember if it was Clarke or Asimov who had a giant computer that was hidden underground in a secret location and needed other computers and technicians to translate between it and people."
I believe you're referring to "The Machine That Won The War", by Asimov. Nobody trusted the data coming in, and everyone secretly cooked it.
Is it Real IT, or to be treated as Fake News with Alternative Facts to become Legendary and Common Main Stream and on Main Street? Are you able to listen and parse to common sense for intelligence, raw novel information? There's more of it about than you realise ....
amanfromMars  …… shooting the breeze on http://www.thedailybell.com/news-analysis/warning-the-media-is-trying-to-kill-you/
You're a tad behind a leading curve with that unfolding news, JJ, but it is encouraging to see it becoming more mainstream. One can only imagine the sorts of new news problems that responsible, and apparently non-accountable status quo systems will be having to deal with in order to try and retain any legitimacy and credibility in the light of burgeoning emerging developments off piste and underground out there in the proprietary intellectual property space which knows of no borders or masters, but it is easy to know of them, for they are not a kept secret ...... Chase Words, Create, Command and Control and Destroy Worlds
IT's a Mad, Mad, Mad, Mad World and you aint really seen nothing yet, for the Greater IntelAIgent Games Show is just at the Beginning of the Times and Spaces Ending Eras of Ignorant Derivatived Darkness and Arrogant Futured Follies. And all are most welcome to Play. And IT's Virtually Free to/in/on all SCADA and Browser Operated Security Systems. And that methinks makes IT priceless and worth more than just billions. And worth practically just as much to remain moth balled and unlaced too, given the death and destruction and disruption Advanced IntelAIgents can so easily cause.
Interesting times and strange zerodays ahead, of that can you be reliably assured.
And you might like to consider and speculate/wonder and ponder on possible and therefore probable future paths with targets and goals for Anonymous/Autonomous/Alien Action/Virtual Machine Proaction/NEUKlearer HyperRadioProActivIT …….. :-) … in clear and present sight of all. And treated as fake news drivers and covers IT with a remarkably effective Unbelievable Stealth.If it's any consolation, those who promote the sick worldviews themselves become a victim of them. ….. Bruce C.
The dark art and ESPecial magic, Bruce C., lies in and with those able to enable a speedy delivery system. You can take it as read that it is a current work in steady progress, albeit not widely know because of the sensitivities which would revolve around the maintenance of security for immunity and impunity of proaction.
> Buzzword bingo to get next grant.
Can you justify your statement in relation to the researcher's blog entry that this article is based on?
It is indeed very well and clearly written, with technical concepts briefly explained in the text, and links offered to a more detailed discussion.
Also, have you ever successfully applied for grants?
"connect the invented languages with English via having the agents communicate with English-speaking agents"
That would be tweets & twitterers, right?
But seriously, am I the only one who is sick and tired of this being called "artificial intelligence"? The current line of research is never going to become I, no matter how many dollars they throw at it. All it is is game theory. Martin Gardner's probably spinning fast enough to power a small town.
There are a few accepted definitions of 'intelligence', but a common denominator is the ability to solve problems. This qualifies. And since it is clearly artificial, then it is apt to call it AI. No one is calling it Artificial Sentience or Artificial Consciousness.
It may be very limited and not general, so one might term it Limited Artificial Intelligence, but hey, it isn't an issue. Why? Because the readership here won't form an impression of its capabilities from its label but rather from a description of what it does and doesn't do.
To answer your question, you are not the only one who holds your view. However, they like you, have yet to provide a definition of 'intelligence’ that explains their opposition to the use of the term AI.
If getting to artificial sentience was equated to traveling from Piccadilly Circus to the Eiffel Tower, we haven't even crossed the Thames yet.
If artificial intelligence is defined only as solving problems, the ABC-1 computer of 1939 or even the first abacus could be seen as "artificial intelligence". I prefer a definition that requires ability to learn about new fields and solve completely different types of problems that they haven't been programmed for. Solving problems is far too weak of a definition. Is a washing machine artificially intelligent? It does solve the problem that I don't want to scrub my clothes against rocks by a stream for half a day each week...
In my mind, a computer that plays chess or Go is not AI, even if it can beat the best humans. If you could take a computer programmed to play Go and it could learn to play heads up no limit poker simply by watching others play it and asking questions that would be AI - wouldn't have to be the best in the world, but it should figure out for itself the concept of seeing what cards are on the table, what cards it holds, and what the odds are of bettering its hand versus the bet require to stay in. Bonus points if it can learn to bluff on its own, by seeing others do it or better yet figure out for itself that would be a way to win more often.
> or better yet figure out for itself that would be a way to win more often.
I seem to recall that Google's Go-playing computer devised its own strategies for winning. No matter.
Thank you for your response. If I understand your second paragraph correctly, you feel that 'intelligence' should be general and flexible, and (I might be mis-reading you here) doing so off its own back in order to satisfy a motivation (curiosity?).
A true AI wouldn't have a motivation, unless "being programmed to learn" counts as a motivation. A computer programmed to play Go or chess doesn't "devise its own strategies". It is programmed for a specific solution space where a potential solution can be measured against other potential solutions and scored according to an algorithm created by humans. It didn't learn what the rules and strategies of chess or Go were, it was programmed with them - "born with them" if you will. It was programmed with the ability to search massive solution spaces a billion times greater than a human could in a lifetime.
It plays like a human who just learned the rules would play, slowly examining possible moves one by one (if that dumb human had an expert player to whisper in his ear with each move he considered with a rating from 1-100 so he could decide to take the highest scoring move) Yes, it can "prune" unproductive paths, but knowing what paths to prune is programmed into it. A Go or chess playing computer isn't the least bit intelligent, it is simply dumb so much faster than we can be that if it has enough cycles its brute force stupidity can beat a human. Even an expert human who wouldn't have to search the entire solution space six moves deep to see a way to force checkmate in six moves, but might only consider 5-10 moves before seeing it.
Your conception about how systems like AlphaGo learn is way, way off the mark.
Famously, Google's DeepMind learned how to play a range of Atari games from an input of raw pixels; in other words, it figured out how to play games just by "watching" them. Link to original paper here.
Obviously needs some kind of counter incentive to make the "language" more flexible.
1 message per complete sentence is bandwidth efficient but not memory efficient.
Which suggests some kind of "penalty" function to encourage them to evolve a balance between verbosity and uniqueness (IE "words") might work.
Biting the hand that feeds IT © 1998–2020