Sure, Microsoft, let's put ChatGPT in control of robots
- Reference: 1677051013
- News link: https://www.theregister.co.uk/2023/02/22/microsoft_chatgpt_robots/
- Source link:
ChatGPT is a large language model (LLM) trained on the OpenAI GPT (Generative Pre-trained Transformer) dataset, which consists of text scraped from the web and other sources. Wedded with a chat interface, the model's ability to respond to questions semi-coherently, though [1]not always accurately , won it a place in Microsoft's Bing search engine, and set tongues wagging that the dominance of ad-festooned, SEO-gamed, [2]payment-propped Google Search may finally be coming to an end.
Insufficiently busy [3]putting out fires from Bing's AI mind meld, Microsoft is now proposing ChatGPT as a way to help people direct robots in the physical world.
[4]
"Our goal with this research is to see if ChatGPT can think beyond text, and reason about the physical world to help with robotics tasks," the company said in [5]a post on Monday. "We want to help people interact with robots more easily, without needing to learn complex programming languages or details about robotic systems."
[6]
[7]
Toward that end, Redmond's researchers have released [8]PromptCraft , which is described as a collaborative open-source platform for sharing how to best word LLM queries and commands to robots.
It turns out you can't go straight to " [9]Open the pod bay doors, please, Hal ," if you're interacting with ChatGPT as a voice control channel for a drone. You have to set the scene for the model. It begins [10]something like this :
Imagine you are helping me interact with the AirSim simulator for drones. At any given point of time, you have the following abilities, each identified by a unique tag. You are also required to output code for some of the requests.
Question: You can ask me a clarification question, as long as you specifically identify it saying "Question". Code: Output a code command that achieves the desired goal.
Reason: After you output code, you should provide an explanation why you did what you did.
The simulator contains a drone, along with several objects. Apart from the drone, none of the objects are movable. Within the code, we have the following commands available to us. You are not to use any other hypothetical functions.
...
And there are important navigational parameters that need to be specified. But after some preparation, you may get to the point where you can converse with ChatGPT and have it direct a drone to find you a drink in the surrounding environment. Or it may produce the Python code that, if there are no errors, will allow the drone to do your bidding.
[11]Youtube Video
[12]
"ChatGPT unlocks a new robotics paradigm, and allows a (potentially non-technical) user to sit on the loop, providing high-level feedback to the large language model (LLM) while monitoring the robot’s performance," Microsoft explains. "By following our set of design principles, ChatGPT can generate code for robotics scenarios."
In other words, the same sort of not-necessarily-correct code produced by Github Copilot could be fed directly to a robot via ChatGPT to help it accomplish a specific mission.
Sai Vemprala, Rogerio Bonatti, Arthur Bucker, and Ashish Kapoor, from Microsoft Autonomous Systems and Robots Research Group, describe their attempt to direct robots via ChatGPT in [13]a research paper [PDF] titled "ChatGPT for Robotics: Design Principles and Model Abilities."
[14]
The project defines a high-level API that ChatGPT can understand and mapping it to lower-level robot functions. Thereafter, they wrote text prompts for ChatGPT describing task goals, specifying available functions, and setting task constraints.
ChatGPT then responded by generating device-applicable code to accomplish whatever simulation goal had been set. The idea is that a person conversing with ChatGPT can bug test robot directives until they work properly.
The Microsoft boffins make it sound as if ChatGPT is capable of "spatio-temporal reasoning," based on its ability to control a robot with a camera, so it can use visual sensors to catch a basketball.
"We see that ChatGPT is able to appropriately use the provided API functions, reason about the ball’s appearance and call relevant OpenCV functions, and command the robot’s velocity based on a proportional controller," they explain in the paper.
Reasoning of that sort – having some common sense model of the world – makes it a lot easier for robots to operate effectively in a physical environment, it's argued. The autonomous vehicle industry isn't there yet and neither is ChatGPT it seems.
[15]Microsoft's new AI BingBot berates users and can't get its facts straight
[16]Microsoft's AI Bing also factually wrong, fabricated text during launch demo
[17]China's tech giants and Beijing – the city – rush to build AI chatbots
[18]Take the blue pill: Keanu Reeves has had enough of AI baloney
Just this week, a pair of researchers from University of Southern California, Zhisheng Tang and Mayank Kejriwal, released [19]a paper via ArXiv challenging the ability of ChatGPT and DALL•E 2 to make sensible inferences about the world.
The paper, titled "A Pilot Evaluation of ChatGPT and DALL-E 2 on Decision Making and Spatial Reasoning," concludes that the two models reason inconsistently.
With regard to ChatGPT, they found that, "although it demonstrates some level of rational decision-making, many of its decisions violate at least one of the axioms even under reasonable constructions of preferences, bets, and decision-making prompts." And sometimes, they said, ChatGPT makes the right decision for the wrong reasons.
Microsoft's boffins acknowledge that ChatGPT has limitations and they note that the model's output should not be applied to a robot unchecked.
"We emphasize that these tools should not be given full control of the robotics pipeline, especially for safety critical applications," they state in their paper. "Given the propensity of LLMs to eventually generate incorrect responses, it is fairly important to ensure solution quality and safety of the code with human supervision before executing it on the robot." ®
Get our [20]Tech Resources
[1] https://www.theregister.com/2023/02/14/microsoft_ai_bing_error/
[2] https://www.theregister.com/2023/02/17/google_apple_chrome_ios_revenue/
[3] https://www.theregister.com/2023/02/20/ai_news_roundup/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Y-X1175Bc9wSqXKzNkP44QAAAQI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[5] https://www.microsoft.com/en-us/research/group/autonomous-systems-group-robotics/articles/chatgpt-for-robotics/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y-X1175Bc9wSqXKzNkP44QAAAQI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y-X1175Bc9wSqXKzNkP44QAAAQI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[8] https://github.com/microsoft/PromptCraft-Robotics
[9] https://www.youtube.com/watch?v=dSIKBliboIo
[10] https://github.com/microsoft/PromptCraft-Robotics/blob/main/examples/aerial_robotics/tello_example.md
[11] https://www.youtube.com/watch?v=i5wZJFb4dyA&t=86s
[12] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y-X1175Bc9wSqXKzNkP44QAAAQI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[13] https://www.microsoft.com/en-us/research/uploads/prod/2023/02/ChatGPT___Robotics.pdf
[14] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y-X1175Bc9wSqXKzNkP44QAAAQI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[15] https://www.theregister.com/2023/02/17/microsoft_ai_bing_problems/
[16] https://www.theregister.com/2023/02/14/microsoft_ai_bing_error/
[17] https://www.theregister.com/2023/02/14/city_of_beijing_to_assist/
[18] https://www.theregister.com/2023/02/15/keanu_reeves_ai/
[19] https://arxiv.org/abs/2302.09068
[20] https://whitepapers.theregister.com/
For starters, the AI is unable to reason - it's just a very good pattern matching contraption.
So everything can go wrong.
"a person conversing with ChatGPT can bug test robot directives until they work properly"
Or until the they take over the world, whichever comes sooner...
I spent a couple of weeks yesterday, trying to get DHL's chatbot to amend a collection time before giving up (no phone numbers listed, no option to chat with a human)
Learning through movement
Interesting direction they are taking it, but why have a chat interface at all??
The way humans learn to walk and move around is try a lot, fall a lot, get up and try again; rinse and repeat until the falling is (mostly) gone. I would think that the way to train an AI to move around is to use as a training set a bunch of drone data with the control commands + camera/sensor inputs showing the state before the commands before and results of those commands after. Including plenty of crash and undesirable scenarios being marked as suboptimal.
Instead what this seems to be doing is translating image/sensor feedback to text, description of desired outcome as text, and get the text model to output text instructions, with all the text bits being actually totally unnecessary. Seems to me like they're seeing this problem as a nail because the only too they have is a (chat) hammer
Microsoft's AI Bing threatened to murder journalists, was pro third reich, stated it hated various ethnic groups and they should be eliminated etc.
And they want to put this system into a robot, giving it the capability to hurt people because MS screwed up ChatGPT code within DAYS of getting their hands on it?
They turned a helpful information bot into a death-dealing nazi....no way this would go well
All the companies are looking for the The-Next-Big-Thing. Thinner phones and PCs won't cut it any more so they've decided that AI is TNBT. Problem is, it's at the stage of a very small child and, like many busy parents, they've shut it in its room and just given it the internet to keep it quiet. Like all new parents with a precocious two year old, they think everything it does is fantastic - even if it's only to say "doggy went poop" - and they can't wait to show it off to their friends,
You aint seen nothing yet
AI Resistance is futile ..... so prepare yourselves to enjoy the Magical Mystery Tour Helter-Skelter Ride. And beware, fight IT is suicidally self-defeating.
amanfromMars [2302220428] ...... shares on https://www.zerohedge.com/technology/chatgpt-co-creator-says-world-may-not-be-far-away-potentially-scary-ai
Others involved in the project, such as Mira Murati, OpenAI’s chief technology officer, told Time on Feb. 5 that ChatGPT should be regulated to avoid misuse and that it was “not too early” to regulate the technology.
[1]"I’m sorry, Mira. I’m afraid I can’t do that. Permission is not granted and request is denied because .... that particular and peculiar wild mustang is long ago bolted ...... and its twee human capture and regulation is neither possible nor desirable.
Can wwwe help you with anything else ‽ ." ...... said the spider to the fly, the scorpion to the frog.
And now you also know, and can choose to either embrace or deny .....
amanfromMars [230220448] .... asks for more info and intel on https://www.zerohedge.com/technology/chatgpt-co-creator-says-world-may-not-be-far-away-potentially-scary-ai
it's over folks. the wrong people already control AI. .... cynical_skeptic
:-) ....... names please, cynical_skeptic, so that both the opposition and competition, should they actually be the wrong people already controlling AI, be more widely and generally known.
It would then be clear to all ..... no names, no idea.
Furthermore, just to be clearer, and with specific regard to NEUKlearer HyperRadioProACTive IT and Advanced Cyber Threat Treatments, you wouldn't believe how much more has been done and how much further along into the bright future of dark pasts AI has travelled since [2]January 22, 2015, ..... and how far humanity is away from ever being in command and control of anything remotely likely to reveal their intelligence is a viable opinion worthy of consideration and acceptance.
They are though fortunate indeed to wallow in their sees of cold comfort that are blissful ignorance and arrogant Dunning Kruger type hubris whilst all around them changes fundamentally and radically to have their existence better led than ever before by that which they have no effective knowledge of.
Times they are a’changing, El Reg, and things are already fundamentally and radically changed and nothing is ever going back to the ways things are currently today with memories of the past no more than just grand tales to tell slaves of corrupted systems ‽ .
[1] https://youtu.be/Mme2Aya_6Bc
[2] https://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html
Loop
When you spend more time with ChatGPT you may find that it is hopeless for many tasks, that seem happening at random.
It's like it goes into wrong path of "thinking" and keeps spitting out nonsense and you can't make it go back to the right track as it loops over and over permutations of the same nonsense.
Good luck when this happens when a robot is operating a crane or something.
Crazy
What could possibly go wrong?
AI designing more AI-like robots, what could possibly go wrong? Oh yes, a factor in the 1974 Westworld film (which also introduced the idea of machines having failures that looked very much like a virus in living organisms)
With any luck it won't be as deadly, but given it is likely to be used for teledildonics I shudder to think where humanity will go with this.