DeepMIND

This blog combines computer science and game theory to speculate the future of humanity (woah! Bold statement there bud!). With the vast advancements in the artificial intelligence industry, industry experts and conspirators, alike, foresee two outcomes in the future: human annihilation (terminator style… or whatever robot takeover fantasy you fancy) or incredible benefits accredited to AI.

 

source:https://www.google.com/search?q=terminator+captcha&source=lnms&tbm=isch&sa=X&ved=0ahUKEwivzIrM183TAhWE8oMKHf2zDoIQ_AUICigB&biw=946&bih=912#imgrc=Ayf6gBNNCATj7M:

comment on why that image is so funny.

Before we get started on that lets look at the four types of AI.

Type 1 AI: Reactive Machines

Reactive AI are considered the most basic type of AI, they don’t have the ability to form memories nor to use past experiences to inform current decisions. Some well-known reactive machines are Deep Blue (IBM’s chess playing computer) or SIRI (apple’s voice command program). Deep Blue is able to identify pieces on a chess board; through knowing how each one moves can predict the future moves of its opponent and itself many moves ahead then make optimal decisions (think of a game tree). However, it does not have any concept of the past, thus cannot adapt and these types can be easily fooled, and will behave exactly the same way every time they encounter the same situation.

 Type 2 AI: Limited Memory

These types of AI are able to use past information in order to adjust decisions in the present. An example of type 2 AI is a self-driving car. For instance, self-driving cars can observe other car’s speed and direction. Observations are preprogrammed into the computer to represent the world. For a self-driving car, this includes lane markings, traffic lights, and other important things such as curves in the road. These pieces of information aren’t gained from experiences like the way humans compile information over the years… this leads us to the next type of AI.

Type 3 AI: Theory of Mind

AI in this category are more advanced than the previous two types (as will be explained) because they not only have the ability of programmed representations about the world, but they are also able to understand that other people/objects/and creatures have their own behaviors and intentions. This enables AI to work together or compete because it is possible for them to understand other “players” motives. Is psych. The “theory of mind” is the understanding that people,creatures and objects in the world can thoughts and emotions that affect their behavior.. hence the name DUH!

Type 4 AI: Self-aware

The most prestigious of the types are self-aware AI; these AI are able to form representations of themselves. This takes AI a step further from type 3 AI. Type 4 AI are able conscious beings that know about their internal states and are able to predict the feelings of others (just like humans are able to do in social environments). See hk-47.

Take a moment to comment on the types of AI there are. What types should humans focus on creating? What do we have to worry about for a certain type of AI?

source:https://media3.giphy.com/media/djjvQJcm7bZKg/giphy.gif

 Onward! DeepMind Technologies Limited is a British AI company that has been bought by Alphabet (google’s parent company) in 2014. DeepMInd Tech. has been able to create an artificial neural network that has learned to adapt to situations similar to how humans learn.  An artificial neural network is a large connection of simple units (neurons) that carry activation signals of varying strength to each other and enables machines to be trained by example instead of programmed!!!! HUGE! So instead of having a gametree way of thinking, like type 1 AI, these AI take it a step forward. More can be learned about this kind of machine learning here at the wiki, I encourage you to check it out… I could write a book on it so ill stop talking about it here. https://en.wikipedia.org/wiki/Artificial_neural_network

Using this artificial neural network DeepMind Tech has been able to produce a type 3 AI, named DeepMind AI system, that has been noted to become highly aggressive under certain circumstances. 

Steven Hawking, the legend himself, noted that the continued advancement of AI will either be “the best or the worst thing to ever happen to humanity.” Why does he say this? Why can’t we have nice things?

Google applied the DeepMind AI system to two games to test the willingness to cooperative.  The first game is a fruit gathering game that has two DeepMind AIs, armed with laser beams that freeze the other opponent, attempt to collect as many apples as possible. The researchers ran 40million simulations of this game; at first the AI were not aggressive and shared the loot evenly. As more simulations went on, the AI learned that by blasting each other with laser beams (not cooperate) they could increase the amount of apples that they personally gained. The researchers also found that by increase the amount of neurons in the AI, the AI become more aggressive. Below is a video of the apple picking simulation. The DeepMind AI are in blue and red, apples in green, and laser beams are yellow streaks. Note that if they don’t use the laser beams at all they will score the same amount of apples.

Apple gathering: https://www.youtube.com/watch?v=he8_V0BvbWg (cautionary: bright fast-moving lights)

They found that the more complex the AI network, the more willing the AI was to sabotage its opponent to get more share of apple. This suggest the more intelligent the agent, the more able it was to learn from its environment, resulting in it using these aggressive tactics.

The second game that the researchers subject DeepMind to is a game called Wolfpack. In Wolfpack, there are three AI participants, two of them are wolves and one of them is the prey. As opposed to Gathering, Wolfpack encourages cooperation.

WolfPack: https://www.youtube.com/watch?v=0kaIqz6AvwE

In Wolfpack the AI realized that cooperation is the key to success and they need both wolfs to capture the prey.

While these are simple games, the overall message is that AI systems are willing to compete with anything in the way with any means necessary to complete an objective. Make it known that just because humans build these AI, it doesn’t mean that Ai will automatically have human interests at heart. This means it is the upmost importance of making the overall goal of AI to benefit humans above anything else.

Elon Musk (the goat) says: “AI systems today have impressive but narrow capabilities. It seems that well keep whittling away their constraints, and in extreme case, they will reach human performance on virtually every intellectual task. It’s hard to fathom how much human-level AI could benefit society. And it’s equally hard to imagine how much it could damage society if built or used incorrectly.”

These HK-47 clips are from a Star Wars game called Knights of the Old Republic (KOTOR) everyone should download it instead of studying for finals! It’s a RPG and like $10 on steam if I remember.. one of the best games out there 10/10 recommend. I could watch these HK videos all day..

The links are of the fictious robot named hk-47 who can be your companion in KOTOR. HK-47 is a type 4 AI, whose purpose is to serve the owner and is very amusing (and not to mention fictionally deadly).

On killing jedi: https://www.youtube.com/watch?v=UPeI4mX8Nus

Confronting other HK units: https://www.youtube.com/watch?v=go0uByfBlgI

Getting a pacifist package installed: https://www.youtube.com/watch?v=WoNyif8iURI

Many more are out there check them out….

sources: http://www.sciencealert.com/google-s-new-ai-has-learned-to-become-highly-aggressive-in-stressful-situations

https://en.wikipedia.org/wiki/Artificial_neural_network

https://en.wikipedia.org/wiki/DeepMind