One more fortnight's gone and little bot aicu-chan climbed her way into the silver mountains. Previously she was stuck in the high bronze plateaus for a long time, but now she pushed further and found her way.
With some improvements to the MCTS algorithm (regularization technique at the rollout/simulation level), it started to play better. Many good reward cards also pushed aicu-chan higher, but it's still limited to level one cards due to the level one summoners.
The performance of aicu-chan surprised me. I didn't expect a deck limited by level 1 starter deck summoners to play at silver II. Well, let's see how aicu-chan keeps playing within the following seasons.
Giveaway
This time the giveaway will again feature a tiny gold foil card. Comment down below, and after seven days I'll distribute a gold spineback wolf :)
Alright, giveaway is closed.
The winner is @bengy :)
Big thanks! It will be a great addition!
Congrats to your successful project.
And thanks for the giveaway
Posted using Partiko Android
Thank you :)
That is pretty decent for a starter deck only. Do you think it is able to manage that consistently over seasons?
Steem handle @bengy!
Thanks. I don't know whether it will be consistent, it's been running for two seasons and just at the start of the second one I changed a major part of the algorithm. That's the same issue with Aicu, for now, they perform pretty consistent :). But only time will tell.
Very inspirational. For sure Aicu and Aicu-chan are better players than me!
Thanks, I'm glad you like what they do. :)
Am I to understand that you're letting your A.I. play? That challenges an ambivalent feeling in me. On the one hand, curiosity about what the AI can bring forward, on the other hand, the question of whether you have marked it as an artificially intelligent player? Since the famous "Go" player was beaten by an AI, I wonder where the direction is going. A world where only AI's play against each other because human opponents have no chance against them? Will the manufacturers make this known? On the other hand, the human play instinct seems unbroken and the gaming industry would probably harm itself by sending AI's into the field ... I'm not informed enough, but there seems to be a strange development on the internet, where suddenly avatars, bots and AIs can't be distinguished from human real people. There are rumours about economic damage and the like. Especially click farms and such obscure stuff.
Since you are a professional and study such things: have you written an article about the above topic or are you planning to do so?
Yes, my AI is playing steem monsters. Whereas the term "my AI" is a bit of a misnomer. I do have two separate programs (aicu the curator (also the name origin of the account) and aicu the SM bot) using various techniques from machine learning, but that's just nitpicking.
Anyway, the SM bot uses the same base algorithm like the Go bot you're referencing. Just without neural networks (working on that, but work is keeping me busy).
The issue with the advances in AI is, that more and more tasks will be assisted or dominated by it. Replacing humans at worst. But I think the possibility of ai assisting and improving workflows and people in general is the more optimistic outlook.
In the game I can't mark the bot as a bot. I'm communicating it openly, but that doesn't change the fact that it's playing and taking rewards from ranked matches etc. For now it's still pretty beatable, but I intend to change that in the future and improve it further. The issue with that is, that others can build such a bot as well. Less sophisticated methods are actually more efficient.
And its pretty much impossible to tell whether a player is a bot or not, if the bot uses the same methods to interact with the game like other players. Unless you know that it's a bot.
That issue just gets aggravated with further improvements in AI and CS in general. And it's not just games which are affected. The latest advances in text generation actually allow people to spread propaganda faster than ever before. Deep fakes on the other hand can deceive people. There have already been cases where someone faked the voice of a CEO and managed to convince a lower manager at the company to send money to a bank account. I haven't fact checked that article, but knowing how people manage things at the companies I worked for, it doesn't sound that unbelievable.
So the economic gain and damage of AI techniques will be tremendous.
And I haven't written an article about it and neither am I an expert in this field. I only have experience in natural language processing, so I can at most comment on text generation, stilometry, voice synthesis, altering etc. But I'd need to read up on it as well. Never looked too much into deceiving people using AI, but the latest developments are giving plenty points of attack to choose from. I won't write an article about stuff like that soon, but in the long run I might do a writeup on fake articles and how they might affect steemit. If I manage to get an article trending using techniques like that or build a following you'll definetly hear from it. Not sure when I get around to experimenting with it though.
Posted using Partiko Android
Makes me think. I think that any human community can basically compensate for criminal acts and it will never be the case that violations of the law or theft can cause fundamental harm to a human community. But this is something different with AI, because they can cause damage in masses ...
Logic dictates that AIs should only communicate with AIs so that the errors you describe cannot happen. However, this also makes strange thoughts ...
Not being able to distinguish between talking to an AI or to a human being always occurs only in remote communication. You could be an AI, couldn't you? LOL.
I've read somewhere that in hospitals, for example, when transferring from one shift to another, it's still best for colleagues to talk face-to-face and orally alongside reading medical records.