Google’s New Robot: Your Personal Butler! 🤖

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. Today, you are going to see how one area  of research can unlock new capabilities   on a seemingly completely different area. We  are now living the advent of amazing AI-based   language models. For instance, OpenAI’s GPT-3  technique is capable of all kinds of wizardry,   for instance, finishing our sentences, or creating  plots, spreadsheets, mathematical formulae,   and many other things. While their Dall-E 2 AI is  capable of generating incredible quality images   from our written descriptions. Even if  they are too specific. Way too specific.    Now note that all this wizardry is possible as  long as we are dealing with text and images. But how about, endowing a real  robot, moving around in the real   world with this kind of understanding of  language. I wonder what that could do? Well, check this out! Scenario 1. This little  robot it uses GPT-3 and other language models,   and it not only understands it, but it  can also use this knowledge to help us.    Don’t believe it? Have a look. For instance,  we can tell it that we spilled the coke,   and ask it how it can help us out? And,  it recommends finding the coke can,   picking it up, going to the trash can,  throwing it out, and bringing us a sponge.    Yes, we will have to do the rest of it ourselves,  but still, wow, good job, little robot. Now the key here is that it not only understands  what we are asking, propose how to help, but,   hold on to your papers, because here comes  Scenario 2. Oh my! It also looks around,   locates the most important objects required,  and now, it knows enough to make recommendations   as to what to do. And of course, not all of  these make sense. Look at that! It can say that   it is very sorry about the mess. Well, thank  you for the emotional support, little robot,   but we also need a little physical help  here. Oh yes, that’s better. And now, look,   it uses its eyes to look around, yes, I have my  hand, well, does it work? Yes it does, great, now,   coke can, trash can, sponge. Hmm…it’s  time to make a cunning plan. Perfect! And, it can also do plenty more. For instance,  if we’ve been reading research papers all day   and we feel a little tired, if we tell  it, it can bring us a water bottle,   hand it to us, and can even bring us an apple.   Now I’d love to see this…be gentle, and! Oh   my! Thank you so much! The amenities at Google  HQ seem to have leveled up to the next level. And, believe it or not, these were  just really simple things, it can do   way better. Look. This is a plan that  requires planning 16 steps ahead,   and it does not get stuck, and doesn’t mess  up too badly anywhere. This one is as close   to a personal butler as it gets. Absolutely  incredible. These are finally real robots that   can help us with real tasks in the real  world. So cool! What a time to be alive! Now, this is a truly amazing paper, but  make no mistake, not even this is perfect.    For instance, the success rate for the planning is  about 70%, and it can properly execute the plans   most of the time, but clearly, not all  the time. The longer term planning results   may also need to be a bit cherry-picked to  get a good one. It doesn’t always succeed. Also, note that all this is played at  10x speed, so, it takes a while. Clearly,   typing the message and waiting for the robot  still takes longer than just doing the work.    However, this is an excellent opportunity  for us to apply the First Law Of Papers,   which says, that research is a process. Do not  look at where we are, will be   two more papers down the line. If OpenAI’s image  generator AI, DALL-E looked like this, and just   a year and a paper later, it looks like this.   Well, just imagine what this will be able to do   just a couple more papers down the line.   And, what do you think? Does this get

your mind going? If you have ideas for cool  applications, let me know in the comments below! Thanks for watching and for your generous  support, and I'll see you next time!