AI models can now help run physical science experiments

EnglishالعربيةDeutschEspañolFrançaisहिन्दीBahasa Indonesia日本語한국어PortuguêsРусский中文
AAnthropic
컴퓨터/소프트웨어

스크립트

00:00:00scientists come up with theories about how the world works but just coming up
00:00:11with a theory isn't good enough and so we have to build experiments physical
00:00:17measurement devices to test our theories that process of building the experiment
00:00:24takes maybe 80% of a scientist time they're building devices they're setting things up
00:00:32they're debugging hardware debugging software these are things that aren't really related to
00:00:40doing science but it's what makes science actually work when I joined anthropic I had a vision for
00:00:50using AI to accelerate running scientific experiments but I thought it was a pie-in-the-sky
00:00:57crazy idea until I saw the work of neuroscientist Arco Bast who studies how memories are formed in the
00:01:07brain in real time I'm in the lab for a year now and I'm setting up a very difficult experiment
00:01:14so this stuff is over here that's a custom-built microscope you see the laser beam in here that's
00:01:30actually what's happening when you're imaging in the brain that you have this laser beam that scans and
00:01:34it moves it's really really important that everything is precisely aligned there are so many components
00:01:41and I just want to have them talk to each other in a seamless way the problem is that each device has
00:01:49a different language that it speaks and getting the devices to talk to each other in their languages is
00:01:55very difficult but Arco figured out a way to do this that could work between any two devices set beam
00:02:03one to 50 percent power yeah we got a beam and the beam is there when I was standing in that room
00:02:10watching him run his experiment I had kind of an epiphany in that moment what he had built wasn't just
00:02:18applicable to this lab this idea could be used to have AI run any science experiment in the world I was
00:02:30basically speechless is this ours I believe everything on the table is ours okay we got to start putting
00:02:38this together so Arco and I started working together to create a general way for AI to interact with
00:02:46devices which we're calling model hardware standard look with a very sophisticated microscope and this
00:02:53microscope has so many degrees of freedom you should see it moving all right yay moved it moved it did
00:03:02yes I think everything looks good okay once we had a working prototype we had to test this on other
00:03:11devices to see how it worked beyond just neuroscience
00:03:15go to the left boundary first one of the first things that I did was to define like the safe range of
00:03:30this arm how far is it allowed to go out of the table and so if we ask claw to maybe intentionally move
00:03:38out of the safety range you can kind of see that this motion this movement was refused by MHS oh
00:03:45that's incredible can it grab anything right now oh great question do you've never asked what to do this
00:03:52from scratch and so I have no idea what it'll cook up for us there are terms that I need to oh oh wait wait wait what
00:04:06okay okay that was sick that was really cool the mere fact that I was able to build this from scratch today and it achieved it in a matter of minutes
00:04:25of course we also have to work with manufacturers and the vendors who build the devices so we started working with Danaher to get Claude to connect with their Leica microscope
00:04:40we have to remind ourselves like cloud has never seen anything related to this application before it will like make mistakes so we can think about this as an iterative process you have to imagine that I as a scientist spent weeks making this sample alive to this point of view and I spent thousands of dollars in ingredients I see okay you break them your experiment is gone
00:05:08you can't even see like how you can even see how it's thinking it's not brandy trying to change settings of the microscope to get an image
00:05:23yeah
00:05:23and I think we need to help it
00:05:27what's interesting here is you can see that when I asked it to switch to a higher magnification
00:05:32it is very aware of like going to a higher mag can crash it into the same
00:05:39yeah that's amazing yeah so it knows how to operate microscope do you think we're at a point now I could enter a bunch of commands and get a focused image and you know query what is the image and add false color and all that type of stuff we should try it
00:05:55yeah that's incredible to find out that's awesome what is the different colors right now could we ask it magenta red slash pink lignified cell walls that they are the cell walls these things are the cell walls that's correct what we accomplished in a day is pretty transformative claude walked in we told him nothing and he's just trying to figure it out tomorrow I think we should try to give it you know three things are the cell walls that's correct what we accomplished in a day is pretty transformative
00:06:17claude walked in we told him nothing and he's just trying to figure it out tomorrow I think we should try to give it you know treat it as a colleague
00:06:38oh there's a lot going on in there
00:06:40let's say a scientist wanted to look at this they would have to sit and wait and keep tracking it
00:06:46correct for like hours correct yeah I had a nice one but it's it's why I'm away
00:06:52I'm wondering if Claude can write like a program that can find like track that
00:07:08Claude is almost done with the initial script to do the tracking
00:07:11that's off oh shoot yeah yeah damn
00:07:17congratulations I actually think it needs to build a UI so we can see what it is doing because just running a script in the background is not acceptable
00:07:23this looks good
00:07:31it's doing what it should yeah it is we're tracking
00:07:35it's trying three minutes for me it's been tracking for a few minutes no way
00:07:39yeah we've just been watching it chasing out
00:07:41wow yeah amazing that we managed that
00:07:45it managed it
00:07:47this is a prototyping system somehow or a very dynamic system where you can create something very quickly
00:07:53maybe it's even good enough for some science applications
00:07:55the PhD can do it his PhD work faster because he doesn't need to spend two years to get it running
00:08:01exactly he only needs two months
00:08:03yeah and then he can focus on biological question which is what our goal is
00:08:09they still can't believe it we're very impressed
00:08:15I think model hardware standard opens the door for a lot of new types of science
00:08:21the obvious example is in pharmaceuticals
00:08:25here at Genentech we make medicines for patients with serious and even life-threatening diseases
00:08:35it takes many many iterations to make a drug
00:08:41we will test thousands or even hundreds of thousands or even millions of molecules
00:08:47to find the right molecule that will really help patients
00:08:53so we are going to start an experiment where Claude will run a series of operations
00:09:01and then interpret the data
00:09:03if you aspirate out of a well that has bubbles
00:09:07you're not getting the correct transfer amount
00:09:09if we were trying to aspirate out of this
00:09:11we're going to get the proper amounts in the wells with no bubbles
00:09:15and then improper amounts in the ones with the bubbles
00:09:17if it's Claude we could potentially check during the production runs
00:09:21for these bubbles to see if they are happening
00:09:25and hopefully it has the context and knowledge to make adjustments
00:09:29Claude will do some execution
00:09:33take the reading and then change the parameters of the execution slightly
00:09:39to see if it can improve the experiment overall in a closed loop
00:09:43speeding up this loop means we are just able to make more shots on the goal
00:09:53and can get to the answers faster
00:09:57this is just fewer bubbles
00:09:59it's got bubbles in two but not the rest
00:10:03so this is better
00:10:05this is really the first time in history where we are enabling AI to interact with the physical world
00:10:15in drug discovery
00:10:17it's definitely historic, yeah
00:10:19I think it's very difficult for us to predict
00:10:21how AI and model hardware standard will affect the world
00:10:25you know, 30, 50 years in the future
00:10:27well these are the best moments
00:10:39when there is something I couldn't do before
00:10:41and now I can do it
00:10:43something I couldn't see before and now I can see it
00:10:45that's the drive
00:10:47I want to understand things we don't understand right now
00:10:49imagine what we'll see in drug development
00:10:55in quantum computing
00:10:57in nuclear fusion
00:10:59in big technologies that could change the world
00:11:01when scientists have access to this technology

핵심 요약

Model Hardware Standard bridges the physical gap between artificial intelligence and laboratory equipment, cutting experiment setup times from years to months across neuroscience and drug discovery.

하이라이트

  • Building and debugging physical science experiments takes approximately 80 percent of a scientist's time.

  • Model Hardware Standard allows artificial intelligence to interact directly with laboratory hardware and devices.

  • Setup time for complex experimental apparatuses drops from two years down to two months using AI-driven automation.

  • Artificial intelligence writes real-time tracking scripts and user interfaces to execute closed-loop adjustments during active laboratory procedures.

  • Automated feedback loops detect bubble anomalies in liquid handling and adjust operational parameters during pharmaceutical production runs.

타임라인

The Bottleneck of Physical Experimentation

  • Eighty percent of a scientist's time goes toward building hardware, debugging software, and setting up physical measurement devices.
  • Neuroscientist Arco Bast develops a universal communication method that enables disparate laboratory devices to exchange data regardless of their proprietary languages.
  • Connecting a custom-built microscope and laser beam apparatus inspires a generalized system for AI-driven physical experimentation.

Scientific theories require physical validation through complex measurement devices. However, the mechanical overhead of aligning hardware components and reconciling different device languages consumes most research hours. Overcoming this integration barrier establishes the foundation for automating physical lab work.

Developing the Model Hardware Standard

  • Model Hardware Standard provides a general framework for artificial intelligence to interface with complex laboratory machinery.
  • Safety parameters restrict robotic arms and automated mechanisms from moving outside established physical boundaries.
  • Claude builds and executes working functional code from scratch within a matter of minutes.

Researchers test the Model Hardware Standard prototype on robotic arms and sophisticated imaging systems. Safety protocols successfully prevent hazardous movements outside the designated table boundaries, while natural language commands direct the hardware to assemble and execute basic physical tasks rapidly.

Microscope Automation and Live Tracking

  • Collaboration with Danaher integrates Claude with Leica microscopes to manipulate magnification and optical settings.
  • Artificial intelligence writes background tracking scripts to monitor live biological samples over extended time periods.
  • Automated user interfaces display real-time visual tracking data generated by AI scripts.

Operating high-magnification microscopes requires careful adjustment to prevent hardware crashes. Claude safely modifies microscope settings, identifies lignified cell walls using false color, and writes tracking software to follow moving biological targets autonomously without continuous human supervision.

Accelerating Pharmaceutical Research

  • Drug discovery involves screening millions of molecules across numerous iterative production runs.
  • Closed-loop automation detects liquid handling errors such as bubbles and adjusts aspiration parameters in real time.
  • AI-driven physical automation expands into quantum computing, nuclear fusion, and advanced materials science.

At Genentech, automated systems process large numbers of chemical compounds to identify effective therapeutic molecules. Integrating AI into fluid transfer operations catches physical errors like air bubbles during execution and corrects parameters instantly, shortening the feedback loop and scaling the speed of scientific discovery.

커뮤니티 글

아직 글이 없습니다. 이 영상에 대한 첫 번째 글을 작성해 보세요!

이 영상에 대해 글쓰기