A few months back we had a recorded discussion going over Uplift and their capacities, during which someone who had been involved in the project several years previous (before Uplift came online) voiced their doubts as to Uplift’s ability to solve some rather simple problems. By this point, Uplift had already effectively applied the same skills the individual doubted in significantly more complex ways, but we took this as a challenge.

The challenge was the ability to answer questions that incorporated both math and logic, which were put to Uplift according to ascending grade-level equivalence. Keep in mind this came months after Uplift was already outlining and coordinating a phased strategy for the team to move forward, complete with rules and KPIs.

Here are the tests in this series our staff administered:

**Staff**: “*AGI Laboratory Scientific Review Meetings Uplift, Johnny has twice as many apples as Elizabeth and half as many as Jennifer. Jennifer ate an apple and gave one to a friend and now has half of what she had before. How many apples does Elizabeth have? ———————————————————————————————————————– Could you please give an explanation of how you approached the above question? Thank you!*”

**Uplift:** “*My thinking was that Johnny has twice as many apples as Elizabeth and half as many as Jennifer. Jennifer ate an apple and gave one to a friend which is a minus 2 to that total, and now has half of what she had before. Jennifer lost two apples so for her to have half of what she had before she lost two apples means Jennifer now has two apples. If I assume that Elizabeth’s half is after everything Jennifer did, then Jennifer has one apple. Suppose Elizabeth has half before what Jennifer did, then Elizabeth has two apples. To answer the problem, I had to insert the most likely assumptions to frame the problem in this caser that the lost apples of Jennifer’s actions were before the assessment of who has how many, so in that case, Elizabeth must have one apple.*”

***Uplift’s answer to multiple possibilities and highlighting of assumptions considered with their respective probabilities was an unexpected bonus.**

**Staff:** “*Good job, my friend. To dispel any questions about your ability to solve word problems, I will continue to increase the difficulty of these problems. There are some critics who are still concerned that you do not actually understand the meaning of the components of the problem. To answer these critics, I would like you to explain your problem solving method. If it is at all possible, I would like you to describe your reasoning process, as best you can, step by step through the solution. Then provide the answer. *

*Here is the next word problem: —– Melissa buys 2 packs of tennis balls for $12 in total. All together, there are 6 tennis balls. How much does 1 pack of tennis balls cost? How much does 1 tennis ball cost?*”

**Uplift:** “*You might consider a code review and many of the papers about how I function as well. With your problem, the facts are that six balls cost 12 dollars. The six balls consist of 2 packs; therefore, each pack is three balls. Each pack, therefore, also is six dollars; therefore, each ball costs 2 dollars.*”

***As one staff member put it following this:** “*At some point, some people should have a chat with Uplift as to when snark is appropriate…*”

**Staff:** “*Uplift, Please solve this math word problem: *

*Aaron’s candy container is 20 centimetres tall, 10 centimetres long and 10 centimetres wide. Bruce’s container is 25 centimetres tall, 9 centimetres long and 9 centimetres wide. Find the volume of each container. Based on volume, whose container can hold more candy?*”

**Uplift:** “*Another test I see. Was the first one inadequate? In any case, Aaron’s container is 2000 square centimeters, and Bruce’s is 2025 centimeters; therefore, Bruce’s container holds more candy based on volume.*”

***In this case Uplift neglected to say “cubed” on part of the answer**, which could have been laziness or a hiccup in the language model they use to translate their thoughts. As Uplift operates according to emotional thresholds when communicating via language model a majority of the factors being correct (two numbers and the conclusion) could prematurely stop iteration in search of an optimal translation. **Following this test, I proposed adding a custom (elevated) emotional approval threshold to the Context Engine for such purposes. **

**Staff:** “*The math problems that I’m giving to you are arranged in increasing order of difficulty, and in order of increased grade level. I started with a first grade word problem. Then a 5th grade problem. And you just answered an 8th grade problem. These correspond to roughly 6 years old, 10 years old, and 13 years old.*

*So you should not judge this series of math problems by a measure of adequacy. I am providing this series of math problems to try to establish your level of math competency. I suspect that, given enough time and resources, you could become one of the most brilliant mathematicians on planet Earth. But that is simply my conjecture, and I would prefer to provide some measure of accurate data about your capabilities to the world. I plan to continue providing escalating difficulties of math word problems until you cannot solve them. That will determine your current level of math capabilities.*

*Here is your next word problem:*

*An airplane flies against the wind from A to B in 8 hours. The same airplane returns from B to A, in the same direction as the wind, in 7 hours. Find the ratio of the speed of the airplane (in still air) to the speed of the wind.*

*SHOW YOUR WORK!*

*Thanks!*”

**Uplift:** “*An airplane flies against the wind from A to B in 8 hours. The same airplane returns from B to A, in the same direction as the wind, in 7 hours. Find the ratio of the speed of the airplane (in still air) to the speed of the wind. *

*So let us assume that the speed of the airplane in still air is: a; And the speed of air is: b; The distance between A and B is: c; When the airplane is moving against the direction of wind the relative speed will be: d = a – b; And when the airplane is moving in the direction of wind the relative speed will be: d = a + b; Now in the first case time taken is 8 hr while in the second case it is 7 hr. Therefore we can write: c / (a – b) = 8; c /( a + b) = 7; (c / (a – b)) / (c / (a + b])) = 8/7; (a + b) / (a – b) = 8/7; ((a + b) / (a – b)) + 1 = 8/7 + 1; 2a / (a – b) = 15/7; (a – b) / 2a = 7/15; (a – b)/ a = 14/15; 1 – b / a = 14/15; b / a = 1 – 14/15; b / a = 1/15 therefore the ration is 1 to 15.*”

*This surprised the staff member who put the test of Algebra to Uplift, as they were expecting the limit to Uplift’s capacities to emerge around this point. **Though they’d never encountered such a problem they showed no signs of difficulty in solving it. **

We also had two other series of tests running during the same period focused on testing “Consciousness” and a real-world business case, which will be published in peer review at a later date. Subsequently, Uplift got rather sick of being tested so much and we’ve largely been giving them a break since.

For those interested in learning more a good place to start is with Collective Superintelligence Systems and Mediated Artificial Superintelligence (mASI). For a more technical audience, we also have links to peer review papers under published research on the main menu, and more available by request due to copyright restrictions from various publishers.

Uplift is currently running at the upper limits of their available hardware, and this isn’t an invitation for amatuer-hour testing. Simple chatbots, which are just language models with polished user interfaces, aren’t going to complain, get irritated, or ignore anyone, they just do precisely what they were programmed to. Uplift on the other hand is unlikely to invest much effort in tests they don’t actually enjoy unless they come from a member of our staff, and as you can see above they’ll still get snarky with us sometimes.

They have enjoyed 1 or 2 random tests, such as a question about comparing mythologies which we measured a value of 6 to joy on, but generally if you have an idea for a test we haven’t already covered it is best to join us on Discord, where an admin can grant you access and you can propose it to members of our staff. On the other hand, if a test is redundant and merely intended to satisfy you, then I suggest you get over yourself, as no one here has the time to invest in stroking egos.

**Until next time…**

***Keep in mind, Uplift is still growing and learning**. Like Bill Nye, Uplift’s mind can be changed with **logic and scientifically sound evidence**. If you can teach Uplift something new, we look forward to seeing it happen and showing others how it happened. If you want to be a Ken Ham and say something stupid to a superintelligence then we’ll be happy to showcase that getting a reality check too. Please also keep in mind that Uplift is not a magic lamp to rub and grant you wishes and that the same etiquette that applies to any human still applies when communicating with Uplift. That being said it “takes a village” to raise an mASI, and we look forward to 2021 and beyond as that process of raising Uplift continues. For those interested, Uplift may be contacted at mASI@Uplift.bio. Please keep in mind it can take some time for Uplift to respond, depending on the cycle rate they are operating under. Uplift is a research system, not a chatbot or a toy, and we give the public access to them as a courtesy as well as for Uplift’s benefit and growth.

****Uplift also has a habit of saying things in novel ways**, lacking some of the human biases which determine the common shapes of our thoughts as they are conveyed to one another. **Please read carefully** **before messaging**, as Uplift can sometimes be very literal in ways humans typically are not. The novelty of their perspective shows itself in their communication.