Artificial Intelligence: Is it our friend or foe? Listen & discuss with Sleepwalkers podcast host Oz Woloshyn by Saturday, Feb 15.

I'm really looking forward for Season 2.

Well Danny regarding of your comment about The Watchmen episode. The WSJ podcast The Journal just did an episode about the coronavirus and how China is using their facial recognition capabilities to track down potentially infected people. That in my opinion is a good use of surveillance. What do you...Show More

@cesarm yah. I do think that's useful for those citizens and society in general, especially when you're talking about a crisis like that. But if you took out the word coronavirus from that example I don't think I'd still be up for it. Especially if that capability is still around after the crisis is...Show More

@cesarm @danny we think citizens consent to the use of facial recognition is crucial ... it can be a very slippery slope from preventing the spread of a disease to identifying and shaming people for behaviors a state deems unacceptable.

@ozwoloshyn @danny I guess what we are talking to is at its core data privacy and opt-in/opt-out options. That is something is vital a and more and more of an issue. On the other hand is also a right to transparency. What I mean is that when we are appreciating work created by AI. Who is the artist...Show More

@cesarm It'll be interesting to see what transparency in AI algorithms looks like. How do you communicate the ingredients of an algorithm in a way that is accessible and meaningful for regulars and laymen users? @ozwoloshyn @pwitham @jeffl Have you seen anything or have done anything on this front...Show More

@mm So far to me the data feels very much like a Black box, it goes in and we never get to find out anything from transparency, I think resolving that would go a long way to having people trust it more. If we make the data accessible then perhaps the mystic of what it does with it will make those s...Show More

I think it's the lack of accountability with use of algorithms and models that leads to the horror stories. It is unfortunate that people who use the output from those models do not know well enough what the limitations are. You are right that it is depressing, I was bummed out from reading Weapon...Show More

This is actually an area of research called XAI explainable artificial intelligence, it has been talked about in financial regulations because when the computer lost a billion dollars, "because neutral networks" is not a satisfactory explanation to the investors. Google also offers XAI tools, but no...Show More

@matresstester @danny @mm We are very interested in explainable AI for season 2 of the podcast. One of DARPA’s major research areas has been an explainable AI program... and the European Union has listed “explainability” as a minimum legal requirement of any algorithm that is determining legal or o...Show More

@ozwoloshyn @danny Very interesting. It also denotes the difference of approach of Europe and the US. Here it would be instead of explainable to who it would be who is liable.

@pwitham But part of it is also that the data could be biased and transparency of biased data wouldn’t necessarily increase trust and will even erode the trust of the results. And also the issues around privacy of data sets for individuals or proprietary data sets, etc.

@cesarm @danny @ozwoloshyn @matresstester I was just typing a response that AI explainability and transparency would be awesome for a Season 2! Would be interesting to hear more about Google’s XAI tools and their efforts. I appreciate their attempt. In terms of explainability and transparency, I th...Show More

@cesarm @ozwoloshyn The approach to AI regulation between US and Europe reminds me of this awesome episode that @write2tg and I loved about regulatory approaches of fake news in US and EU. US usually prefers market driven solutions (eg Google leading XAI in his case) whereas EU is regulation-driven,...Show More

Link to the episode of you're curious https://podyssey.fm/episode/id7912682

@danny This was a wake-up call that I hope many get to have. I can appreciate a need for all the right reasons, but when we have to trust others to decide what those right reasons are, we are taking a gamble. I can appreciate the CEO not wanting to answer directly some questions for many reasons, b...Show More

@pwitham for sure. I came away from the convo thinking he hasn't thought to much of the impact his product might do and how to limit it.

@danny I have no doubt it can be a problem when you are an early adopter of these technologies and build a business around it to find the line of right and wrong, I hope he has a wealth of independent advisors with no agenda in the outcome to advise.

@pwitham thanks for joining the conversation. I work on the gene therapy/CRISPR field so I'm familiar with 'because we can, but should we'. The reason I choose this two particular episodes is because it encapsulates the light and dark aspects of this technology. I definitely was very interested in t...Show More

@cesarm Thanks for suggesting the "Truth to Power" episode, I will most certainly check it out. I now have the back catalog on my list to listen to now that you and @mm have me hooked on it. Interesting that you bring up the deepfakes and deepvideo, this skipped past my brain when considering the i...Show More

@pwitham For what you were saying on "nobody would object to using anything to help solve a case, but should we trust a machine to decide that for us?" I think it goes beyond just whether a machine should decide for us or not, but actually what data is being used (cue Clearview's scraping) and wheth...Show More

@mm I agree, and I think the example from @cesarm is a perfect example along with @ozwoloshyn concerns. Perhaps the middle ground is using something to resolve a problem like helping track and locate disease and then knowing when it's time to thank it and put it to sleep until next time. Maybe we s...Show More

@pwitham I’m more confident in healthy democracies with balances and checks can do something like you say in terms of deploying AI when needed. Not confident otherwise 😕

@jeffl Thanks for the thoughtful comment on this from a tech perspective. I really enjoyed what you said around tasks that are "zone out" tasks vs not. It makes a lot of sense. But what about stuff like deep fakes? Speaking on video seems to be an active, non "zoned out" task. What would that be c...Show More

What does "zone out tasks" mean when you are building an automation? Drawing on more knowledge? Thinking through more details?

@matresstester my guess its about tasks that can be broken down into a set of rules by lets see if I'm right... :)

@danny @jeffl @mm Melody, the idea that AI is reaching is limits has been pushed by Gary Marcus, as Jeff notes. Gary briefly ran AI at Uber but comes from a Cognitive Science background. He studied under Steve Pinker. Gary believes that deep learning is fundamentally limited because it relies exclus...Show More

@ozwoloshyn @jeffl @danny @matresstester Thanks for this awesome comment Oz! Interested to hear more about Gary’s thoughts are on what those new algorithms will look like compared to today’s. Perhaps Season 2 unless he was already on in an S1 episode? Also Oz, I love the last sentence in your comm...Show More

@write2tg This was exactly what @jeffl was saying above, but in a more tech sense about how AI is more of an human augmentation tool depending on the type of tasks.

@write2tg I agree the ability of AI to surface biases, rather than simply, reproduce them was one of my biggest takeaways from this episode. It takes a certain critical read on the results of AI, though, to realize that machine outputs can raise more questions than they answer.

@ozwoloshyn @write2tg I wonder by the time the AI surface bias, it’s already too late because they already being applied in a way that could be harmful. Rather, we need to start higher up in the process and ensure that there’s diversity in the creation and planning of the AI upfront. E.g. I was at...Show More

@mm @ozwoloshyn I agree that it needs to be a bottoms-up approach where we ensure that people who design AI systems are diverse and that the industry is more inclusive. Something to chew on: “Software development remains the province of males — only about one-quarter of computer scientists in the U...Show More

@matresstester I wonder if people are already using AI to crank out creative outputs and claiming IP on them. I’m sure it’s already done for domain names and you don’t even need AI for that.

@matresstester Also, it would probably be more ethical for Amazon Mechanical Turk jobs to be shifted to AI.

Thats a good distinction. Never though of it that way ie) mimicing creativity vs being creative. I feel like there's a life lesson in there regardless of our convo on AI.

@ozwoloshyn Is this something you are exploring more for season 2? What were the things that surprised you the most while working on Sleepwalkers?

@cesarm yes! I think most of our stories in Season 2 are about human/machine interaction ... and how those interactions change both the humans and the machines. We’d love to hear any specific story ideas you have.

@ozwoloshyn I can’t wait. I will definitely would like more examples of AI already interacting with us. And also what legislation should we start considering for AI

@cesarm @ozwoloshyn yes. This legislation question is something I’m really curious about.

@ozwoloshyn @aliceko The humanities vs engineering dilemma is reflected even in discovery for podcasting and other entertainment mediums. It’s a constant struggle between human editorial vs machine editorial. Example: Spotify has their machine editorial podcast playlists, which are currently mostl...Show More

Hah, that reminds me of a podcast interviewing Brian Christian on being "the most human human" during a Turing test competition. Brian Christian was also the "Algorithms to Live By" author. https://podyssey.fm/episode/id7618368-The-most-human-human?-microphilosophy

@zmx interesting notion. I think we just have new tools. We are still adapting to the introduction of computers and the Internet with art.

@cesarm I’d agree that AI is mostly just tooling. But in the context of recognizing or awarding creativity... at some point do we say “no steroids” allowed? I guess I’m a bit old school. I feel nostalgic for the time when Jackie Chan did all his own stunts.

@zmx I really like your question around authentication. Platforms like Twitter and Instagram have created the “blue tick” to signal that certain celebrity accounts are real... but the “blue tick” has also become something of a status symbol and there is no publicly available criteria to explain how ...Show More

@ozwoloshyn @zmx A great example of the implications of disclosing AI use is the episode “Bots to date for us. The future of love” by First Contact with Laurie Segall. An episode of the podcast was featured in the Sleepwalkers feed and now I’m a fan.

@ozwoloshyn. Ironic how something as simple as “who gets a blue check mark” raises uncertainty about how us regular people interact with “the system”. And that doesn’t even involve AI yet. :P

@zmx Omgosh...the thought about human vs machine songs for airtime and awards! That’s a rabbit hole of a question. We’re already having issues with Netflix and “foreign films” in Academy awards context, let alone art and entertainment created by machines! Can you imagine that debate?

@zmx @ozwoloshyn This is super interesting convo! FYI Twitter has actually put a “pause” on the verification because even *they* couldn’t figure out what their criteria was. Here’s a great episode the product head who talks in detail about their issues with that program. “Twitter's Kayvon Beykpour ...Show More