47 Matching Annotations
  1. Jan 2022
    1. bridge the subsymbolic methods

      Not wholly convinced

    2. Neural methods for NER offer a key advance, because they can distinguish between different senses of words based on context

      Distinguishing the topics is good: something else we've talked about with Michael is pulling together hierarchical models of documents and domains, so that we know which topics relate to which; recommendations could then be made on that basis. BTW, Zoe Corneli has done some recent work with recommendations in the shopping domain (Amazon Shopping).

    3. we expect to be able to reliably link symbolic identifiers to their definitions, an entirely novel advance

      This is still part of our current thinking: however I'd also say that in addition to "novelty" we should talk about why this matters. What is it about linking symbolic identifiers that improves our understanding of documents? I think a core concept here might be that of "fingerprint databases". Understanding symbolic text (or other text in the formal register) should help us build better fingerprints. https://arxiv.org/abs/1304.3866

    4. six month project

      This was very helpful for formulating something with a concrete focus; however, 18 or 36 months would allow us to do quite a bit more.

    5. Can AI be used to help make this technical material easier to access, understand, and apply?

      In subsequent thinking about this project I've been looking a bit more broadly. Maybe it's better to come back to this core focus on Arxiv.

      The vigorous growth and development of Science, Technology, Engineering and Mathematics (STEM) begins to suggest a radical idea: to treat STEM itself as a source of big data...

  2. Oct 2020
    1. AI for Good

      Related to Ethical AI

    2. while there are many ways to apply AI within CI [3], the present project moves in the direction of building computational models of CI itself

      There are a number of interesting feedback loops on this: for example how do computational models of social behavior but also increasing literacy with modelling, helping people learn how to access information. Keeping in mind that most people will have gone down path-specific learning & development trajectories that make grappling with complex issues difficult!

    1. Can autonomous computational agents build an explicit, functional, model of the knowledge and epistemic processes that underlie a large technical corpus written by many authors?

      Build an Agent-based "simulation" of Stack Overflow.

    2. The LEAPQA project will use artificial intelligence to create an online learning support tool for technical training in computer science and mathematics.

      Turn online open source materials into interactive tutorial.

    3. “If you can’t solve a problem, then there is an easier problem you can solve: find it”

      There's a social variant: If you can't solve a problem, there's someone out there who has solved it. Find them.

    1. Agents writing institutions

      Maybe this also relates to 'AI planning' (or, anyway, writing plans)

    2. Publication: IJCAI

      One nice thing about this plan is a "divide and conquer" strategy towards having a reasonably specific output after 1 year.

    3. Time off and plan Year 3

      A "towards" way of thinking that's sufficiently expansive is motivating: this needs to be balanced against doable tasks & achievements.

    4. Questions are similarly rated.

      We can also consider a new repository oriented around questions.

      • E.g., "what is the evidence for climate change?" (Org)
      • Link out to other files that perform analyses of existing data
      • "What is the best measurement we have over time of carbon in the atmosphere?" (this isn't the usual starting point b/c people wouldn't think of this)
      • When do I encounter something that's worth doing some computations about (an Org Roam network of computable objects/processes)

      Potentially deploy with a reboot of Dewey's "laboratory school" (they did cooking...).

    5. Agents will gain points for asking and answering questions about Stack Overflow content.

      If we're interested in understanding social & economic dimensions of things then we should also be thinking about what agent based models can tell us... and what about other things?

      • Seeing ourselves as a potential competitor to fivethirtyeight.
    6. Active Inference bootcamp

      This is the second upskilling task, thinking about how to design agents that learn. Beren Millidge has been working on this stuff: could any of that work be taken over? Do we want to explore other possible.

    7. ML/NLP bootcamp

      This would be the first key bit of upskilling for me, though in principle I could work with people who are already working on this sort of thing, and that are interested in teaming up on a piece of work here.

    8. Milestones and deliverables for first 24 months of the project.

      This was something I proposed for "just me, for 2 years", so this is something that I should consider again now, as I pick up a 2 year contract with Oxford Brookes.

    1. Our long-term vision is computational intelligence based on collective intelligence.

      Though, given the 1.5° C buffer, it may not be very suitable to have a "long term view" that doesn't take account of climate change & human adaptation. So, maybe this needs to be re-jigged around a "why" that is much more pressing than just the hedonistic interest in learning stuff. Furthermore part of an answer to this concern is going to be through "social networks" not just in "scientific computing."

  3. May 2020
    1. To double down on this agenda our aim will be to build the company around open source software.

      From the RM Unger whitepaper:

      "The more one knows and discovers, the easier it is to make the next discovery. If the process of production can be organised on a model of scientific inquiry and experimentalism, innovation can stop being episodic and become permanent. Continuous innovation undermines the basis for the constraint of diminishing marginal returns."

    1. knowledge economy

      Worth having a look at the whitepaper produced by R.M. Unger for Nesta, "Imagination unleashed: Democratising the knowledge economy"

  4. Apr 2020
    1. Dame Wendy Hall and Jérôme Pesenti conclude that “AI could positively affect every area of STEM education”

      I worked briefly in the SOCIAM consortium with Dame Wendy Hall. I wonder if some of the stakeholders in the AI report mentioned here could be approached to ask how they are sourcing technical talent. Benevolent AI would be one interesting possibility.

      A first step could be to follow up with the Southampton folk about the Innovate UK proposal.

    2. [33] “Uncovering the Dynamics of Crowdlearning and the Value of Knowledge”. Proc. Tenth ACM International Conference on Web Search and Data Mining. 2017 [34] “Emergent Complexity via Multi-Agent Competition”. arXiv preprint arXiv:1710.03748, 2017

      These look like particularly worthwhile references.

    3. Universities UK

      Universities are a primary incumbent for any kind of skill sourcing! https://www.universitiesuk.ac.uk/

    4. Wolfram Research

      Another possible speculative angle. They will have a broad interest in sourcing technical talent, not just programmers but also mathematicians.

    5. developing new business models for education using open source software

      London-based experts in this aspect of things include Canonical.

    6. Elsevier

      It's a bit of a strange angle but at the level of a brainstorm we could look at how publishers source technical talent. They are probably not the biggest employer of programmers out there, but it would be a non-trivial number.

    7. Building our Industrial Strategy

      "The aim of the Industrial Strategy is to boost productivity by backing businesses to create good jobs and increase the earning power of people throughout the UK with investment in skills, industries and infrastructure."

    8. IBM

      I have a couple of rather remote links to people who work at IBM, I wonder how they go about sourcing talent there.

    1. Simulating Developer Communities

      I found some earlier work, though it's not very fine grained.

      • Modeling the Free/Open Source Software Community: A Quantitative Investigation


      • Agent-based Simulation of Open Source Software Evolution
    1. A short bio and link to LinkedIn would be perfect.

      A link to Github would perhaps be more relevant for many developers. Indeed, wouldn't a good starting place be to crawl Github and try to recruit people whose profiles look suitable?

    1. The real test of meaningfulness will come in the use of the extracted information, via evaluation of agent performance on both synthetic tasks and human-in-the-loop applications.

      This sentence kicks off a paragraph with a grab-bag of different methods: all interesting, but from different domains. As such they seem likely to confuse the reader. Don't get me wrong, TPGs, PP/AI, and IAD are all really cool, but their application here remains speculative! The actual problem being addressed exists at a higher level of abstraction.

    2. “complex feed-backs present between individuals and their environments”

      Clearly this wasn't a watertight case for getting the job. Nevertheless I think there are interesting connections between this way of thinking and the Andy Clark way of thinking about niche construction.

    3. Helena Miton’s work on “the role of institutions in generating and transmitting technical knowledge and practices” would find concrete analogues within the work I have proposed.

      I wonder if we could set up an interview to talk about this!

    4. koans

      As people learn programming, what else do they need besides koans?

    5. In order for agents to explore this domain, they will need stepping stones, starting from the simplest possible tasks and growing in complexity.

      This is a worthwhile principle to keep in mind.

    1. We're beginning by marking up these old proposals with new commentary.

      Annotations are via the Hypothes.is sidebar. Anyone is welcome to add them!

    1. rationale

      Instead of a Problem, Solution, Rationale triple, the Y Combinator folks talk about a Problem, Solution, Insight triple. The insight concerns unfair advantages related to growth. It could come from across several "suits": Founders, Market, Product, Acquisition, and Monopoly.

      Whereas the rationale concerns repeatable patterns, startup ideas concern more time-, location- and context-specific opportunities. So, OK, there are some differences: but it seems to me that the common features between these two design languages is worth keeping in mind!

      The breakdown across different "suits" reminds me also of the multiple capitals theory used by XinX.

    2. We can build computational models of social processes and research heuristics using a formal variant of the design pattern methodology

      What I think I have only just realised today is that the process based material could be combined with Monocl and properly formalised inside of category theory.

    3. The latter will draw on direct observation, interviews, “instrumentation” of the social media accounts of researchers who agree to participate in the study, and software integration work as relevant.

      This epistemic aspect is preserved in many of the proposals and seems to be a key aspect of the offering.

    4. enrich the inquiry with both technical and “common sense” features

      I think the "impact" again falls flat in this paragraph.

      The last bit here harkens back to a disagreement that I had with Bob Boyer at UT Austin circa 2003. I felt it would be worthwhile to model mathematics in more common sense terms, but he did not see the value of that pursuit and wanted me to work on a project more oriented towards theorem proving. But all of the interest (and difficulty) of that proposal is implicit here.

      Naturally people will be interested in the adventure but once people begin to grasp the technical ideas they ask "Is it possible? What's really required? What are the stepping stones or proof points?"

    5. a content-oriented model of technical domains, which will be directly useful for education and research

      It's as if the big ideas here are buried (perhaps because I didn't want to be too bold, but in that case it's strange to write a paragraph about 'importance')! To be clear, if it actually worked this idea would transform the way we do knowledge work. That's a bit more bold! The 2019 whitepaper "Imagination Unleashed: Democratising the Knowledge Economy" might say some of this better than I can, even now! (I have only had a look so far.)

    6. The next step in understanding the relationship between content and process could be made using simulation studies

      This should probably have been pulled to the top of the document, since it is an overall thesis for the proposal.

    7. One example application is a recommender system that shows questions and answers from Stack Exchange to a programmer at work.

      I think this is referring "Seahawk: Stack Overflow in the IDE" by Luca Ponzanelli et al. (https://lucaponzanelli.gitlab.io/).

    8. The people who visit the library can teach you where the data that surrounds you comes from.

      This is basically the thesis statement of this paragraph, embedded here in the middle.

    9. ContentMine

      http://contentmine.org/ "We offer a broad range of text mining services for small, medium size and large projects. Our mission is to give researchers an easy-to-use open source text mining code allowing them to find, download, analyse and extract knowledge from academic papers."

    1. Admas Kanyagia

      It would be great to talk with her about the project of building a 'talent incubator' for engineers!