Talking Points<\/p>\n
In July 2017, when DeepMind opened its first overseas research lab in Edmonton, it was seen as recognition of Alberta\u2019s AI expertise. In January 2023, when DeepMind closed its Alberta office, it was seen by some as a blow. Two and a half years later, Edmonton\u2019s AI ecosystem is thriving.
\nFormer researchers at DeepMind Alberta have launched startups like Artificial Agency and RL Core, which now employ more people between them than the Google AI lab ever did. Ex-staff are also advancing reinforcement learning techniques, which were pioneered in Edmonton, at industrial labs and research institutions in the city.<\/p>\n
DeepMind had been a dream destination for students like Kearney working on reinforcement learning, a field of AI research that trains systems through trial and error. \u201cI wanted to go to this place, DeepMind, and I wanted to work on solving intelligence,\u201d she says. Now, DeepMind was ditching Alberta.<\/p>\n
It wasn\u2019t the first time the province\u2019s AI community had faced doubters. Alberta has always taken its own path to solving intelligence. While other AI research centres focused<\/a> on more logic-based methods or brain-like neural networks, many researchers at the University of Alberta\u2019s computing science department worked on fringe disciplines like reinforcement learning. DeepMind\u2019s arrival in Edmonton helped bring attention and acclaim to that unique strand of AI research. So its departure was taken as a blow.\u00a0<\/p>\n Two and a half years later, Edmonton\u2019s AI ecosystem is thriving. Researchers who stayed when DeepMind left are continuing to advance the science of reinforcement learning at UAlberta and at a new lab they\u2019ve helped open. Other former staff have launched AI startups, like Kearney\u2019s Artificial Agency. Still others are working on AI for multinationals like Sony. Meanwhile, a new cohort of students and researchers are dreaming different AI dreams.<\/p>\n \t\t\t\t\tRelated Articles<\/p>\n \t \t \n\tBy DeepMind\u2019s arrival helped open up the horizons for AI in the Prairies, but so has its departure. Richard Sutton\u2014the reinforcement learning sage who co-led DeepMind\u2019s Alberta office\u2014predicted this. The firm\u2019s decision to exit Edmonton \u201cis really worse for them than it is for us,\u201d he told<\/a> The Logic days after the decision was announced. \u201cWe are freed up,\u201d he said, to \u201cmake new things.\u201d <\/p>\n In the early 2000s, Edmonton was one of a few places in the world where deep networks of researchers were trying to make AI actually work.\u00a0<\/p>\n UAlberta was \u201cthe best place in the world\u201d for reinforcement learning, says Marc Bellemare, who left Montreal for Edmonton in 2007 to work on his PhD in the field.<\/p>\n At the time, reinforcement learning was known as \u201cthe discipline that doesn\u2019t work,\u201d Bellemare says. As a result, few AI researchers bothered with it. Sutton and some of the school\u2019s other professors, however, believed that it would eventually work. By the time Bellemare arrived, they had formed a lab of 30 or so people within the Alberta Ingenuity Centre for Machine Learning (AICML), to try and crack the reinforcement learning code.<\/p>\n By the end of the 2000s, Edmonton was \u201cone of the most prolific places on the face of the earth in terms of graduating AI students,\u201d says Cameron Schuler, the organization\u2019s executive director from 2010 to 2018. Those graduates didn\u2019t have many places to go, though. Only a few professor or postdoctoral positions opened up at universities each year, and corporate Alberta \u201cwasn\u2019t overly interested in AI at the time,\u201d he says.\u00a0<\/p>\n Some from the Edmonton AI school went to Silicon Valley to work as regular software engineers. Others left tech altogether. \u201cWe were still in the AI winter,\u201d says Brian Tanner, who studied at UAlberta between 2003 and 2010. He never finished his PhD, instead moving back to his hometown of Winnipeg to run a fire-safety company.\u00a0<\/p>\n Other UAlberta graduates found work at DeepMind, which at the time was a little-known AI startup based in London. Canadians were at the heart of DeepMind\u2019s early work, and UAlberta alumni formed its reinforcement learning core. Bellemare arrived in September 2013, joining the likes of David Silver, whose doctorate Sutton supervised, and Vlad Mnih, a UAlberta master\u2019s student who\u2019d gone on to a PhD at the University of Toronto under neural networks pioneer<\/a> Geoffrey Hinton.<\/p>\n In those early days, the AI community was skeptical of DeepMind\u2019s work, Tanner says. Many predicted the startup would be \u201ca money pit\u201d and \u201ca total waste of time.\u201d In 2012, when Silver called Tanner to ask if he\u2019d be interested in a job at DeepMind in London, Tanner said no.<\/p>\n Then the AI winter began to thaw. In December 2012, Hinton and two of his students in Toronto presented<\/a> proof that a neural network could significantly beat other AI systems at classifying images.<\/p>\n In London, DeepMind was also generating buzz\u2014and UAlberta alumni were playing a key role. Mnih led a team that combined Alberta reinforcement learning and Toronto neural networks into a new<\/a> kind of AI system that learned to play<\/a> classic Atari 2600 games by \u201cwatching\u201d the screen. It got a lot of attention. \u201cWhen people see a game like Space Invaders or Pong or Breakout, they know it\u2014they understand it,\u201d says Bellemare, who played a key role on the project. <\/p>\n \t\t \tFrom left, Go player Lee Sedol, DeepMind CEO Demis Hassabis and the firm\u2019s reinforcement learning lead David Silver speak in Seoul during Sedol\u2019s 2016 exhibition series with DeepMind\u2019s AlphaGo AI system. Photo: Kim Min-Hee, Pool\/Getty Images\t<\/p>\n DeepMind\u2019s next AI trick also owed a lot to UAlberta. In March 2016, its AlphaGo system beat<\/a> Lee Sedol, the best human in the world at the Chinese strategy game Go. Silver, DeepMind\u2019s reinforcement learning head, had been working on Go-playing AIs since his time in Edmonton.\u00a0<\/p>\n Back in Edmonton the next year, the Alberta Machine Intelligence Institute (Amii)\u2014a rebranded AICML\u2014was chosen as one of three research hubs to deliver<\/a> the federal government\u2019s new $125 million Pan-Canadian AI Strategy. Amii was also trying to get industry more interested in AI, encouraging big businesses like RBC to open<\/a> labs in Edmonton. By early 2017, the institute had \u201ca queue of about 175 companies wanting to work with us,\u201d Schuler says.\u00a0<\/p>\n At the same time, Sutton had been talking to two other research leaders at UAlberta, Michael Bowling and Patrick Pilarski, about building closer ties to the private sector. \u201cWe are a world-class research place,\u201d Sutton says they thought at the time. \u201cWhy don\u2019t we do some world-class company building?\u201d\u00a0<\/p>\n DeepMind liked what it saw in Alberta. In July 2017, Hassabis announced<\/a> the firm would open a research lab in Edmonton to pursue \u201ccore scientific research,\u201d led by Sutton, Bowling and Pilarski.\u00a0<\/p>\n The company\u2019s arrival validated Edmonton\u2019s underappreciated AI expertise, according to people in the industry, and was reward for Alberta\u2019s longstanding investment in the technology. \u201cIt definitely gave some recognition to the area,\u201d says Schuler, now chief commercialization officer at Toronto\u2019s Vector Institute.\u00a0<\/p>\n DeepMind also helped deepen Alberta\u2019s AI talent pool by keeping top researchers in the province and bringing others back. In early 2017, Finbarr Timbers, an Edmonton-born and -raised data scientist, had been applying for jobs in Silicon Valley. Then DeepMind Alberta opened its doors. \u201cThis was the dream job, and I could get in without having to leave my home city,\u201d says Timbers, who joined the lab in November 2017 as one of its first research engineers.<\/p>\n DeepMind\u2019s arrival in Alberta also brought Tanner back. Some six years after turning down DeepMind in London to stay in Winnipeg, he sold his fire-safety business and moved back to Edmonton to join the firm\u2019s new office as a research developer. It was, he recalls thinking, a \u201csecond shot at a once-in-a-lifetime opportunity.\u201d<\/p>\n \t\t \tFormer DeepMind Alberta colleagues Alex Kearney, left, and Brian Tanner co-founded Artificial Agency, a startup that makes AI tools for game developers. Photo: Larry Wong for The Logic\t<\/p>\n Inside DeepMind Alberta, the new hires worked in groups under Sutton to try and make fundamental advances in reinforcement learning, and under Bowling on systems to master more kinds of games. The office eventually grew from 10 to 27 people. While DeepMind\u2019s London headquarters worked on bigger AI applications that required huge processing power, the Edmonton lab focused on earlier-stage AI research, Tanner says. \u201cWe were a very ideas-driven team.\u201d<\/p>\n Former DeepMind staff say the firm had access to larger quantities of compute, and a more organized and experienced team than the academic labs in which they\u2019d previously worked. \u201cWhen industry is good, it kind of can\u2019t be beat,\u201d says Sutton, thanks to an alluring mix of more resources, good pay and fewer distractions. In addition to their research, star professors in a university computer science department still have to teach and apply for grants, and their academic labs are filled with students still honing their skills.\u00a0<\/p>\n DeepMind, by contrast, could afford to bring together dozens of already-trained researchers to just do science, says Bellemare, now a co-founder and chief scientific officer at Montreal-based Reliant AI, which makes<\/a> AI tools for pharmaceutical firms.\u00a0<\/p>\n Abhishek Naik was among those drawn to Alberta by a mix of Sutton and DeepMind. Under Sutton\u2019s supervision, his doctoral thesis focused on developing systems that continued learning through their lifetimes, like animals are hardwired to do. To test his algorithms, Naik worked with engineers at DeepMind including Tanner to create an open-source library<\/a> of never-ending challenges. \u201cI had some vision of what it could be,\u201d he says\u2014and DeepMind helped make it real.<\/p>\n DeepMind Alberta had quickly become a default dream for aspiring AI researchers in Edmonton. With DeepMind gone, they would need to make other plans.<\/p>\n In November 2022, OpenAI released ChatGPT. It sent<\/a> Google, and DeepMind, into turmoil.<\/p>\n Researchers at Google Brain had invented<\/a> the technique behind the LLMs that powered ChatGPT five years earlier only to dawdle on rolling out generative AI products. Now Google had to play catch-up.\u00a0<\/p>\n ChatGPT was \u201cmet with maybe more than healthy skepticism\u201d inside DeepMind broadly and the Edmonton office specifically, Tanner says. Many researchers in Edmonton shared the view<\/a> that generative systems were just regurgitating the material on which they had been trained. DeepMind Alberta hadn\u2019t been working on LLMs, but Tanner drafted a proposal to explore how much they might benefit from reinforcement learning.\u00a0<\/p>\n Quite a bit, it turns out. Today, leading AI labs use reinforcement learning to fine-tune their LLMs to produce better results, while the most advanced reasoning models use the technique to play out different solutions for tasks and problems, and pick the best one.<\/p>\n \t\t \tThe Alberta Machine Intelligence Institute traces its history back to 2002, when the provincial government and University of Alberta jointly established a new AI centre. Photo: Artur Widak\/NurPhoto via Getty Images\t<\/p>\n DeepMind Alberta\u2019s closure in January 2023 meant Tanner\u2019s project never got going. The firm told staff it was shutting the Edmonton lab because<\/a> it wasn\u2019t housed in an existing Google building like the unit\u2019s other sites. Several former DeepMind Alberta employees said the firm handled the closure well, offering researchers and engineers positions at other DeepMind offices in Canada or the U.K. if they wanted to move.\u00a0<\/p>\n While they considered those options, others in Edmonton\u2019s tight-knit tech industry made clear they had alternatives. AI fund Flying Fish Ventures held office hours at Amii for a dozen or so former DeepMind staff who were considering working at or launching startups. It was a chance \u201cto be helpful, but also maybe be part of their next step,\u201d says principal Tiffany Linke-Boyko.\u00a0<\/p>\n Seattle-based Flying Fish had hired Linke-Boyko, a former CEO of Startup Edmonton, in April 2022 to run a new Edmonton outpost because of the city\u2019s AI talent and expertise in reinforcement learning. DeepMind\u2019s presence had helped reinforce the VC firm\u2019s belief in the city. \u201cThe ironic thing is that, as investors, we\u2019re always like, \u2018When are people leaving these places?\u2019\u201d says Linke-Boyko. Suddenly, many were seriously thinking about leaving DeepMind.<\/p>\n Kearney hosted gatherings at coffee shops in Edmonton where DeepMind staff discussed the opportunities they might pursue. Generative AI was clearly going to change the tech sector, and Kearney was increasingly drawn to the idea of working with a small, focused team to build a startup. \u201cIt just felt like the right time to take a risk and do something new,\u201d she says. So she did.<\/p>\n Almost all the former staff of DeepMind Alberta are still in Edmonton today.<\/p>\n Kearney and Tanner, along with two other DeepMind Alberta colleagues, co-founded Artificial Agency in March 2023. The startup is developing AI models and tools that game developers can use to bring their work to life.\u00a0<\/p>\n Artificial Agency\u2019s game engine launched<\/a> in July. The company claims it can make non-playable characters more realistic than ever and turn wooden in-game interactions into complex conversations.<\/p>\n The firm\u2019s technology also lets developers add \u201cgame directors\u201d that adapt the game in real-time based on players\u2019 actions. The tool has full control over the world\u2014\u201calmost like a little God,\u201d says CEO Tanner. Artificial Agency claims studios have already shown interest in using its technology in games under development.<\/p>\n Like so many of the field\u2019s best minds, Kearney is focusing on games because they make the esoteric science of reinforcement learning more understandable to the public. \u201cWe all play,\u201d says Kearney. \u201cThat\u2019s tangible, anyone can get that.\u201d<\/p>\n Kearney, who is Artificial Agency\u2019s head of agents, also wants to inspire more startup founders in Edmonton. The firm now employs 25 people, on par with DeepMind Alberta at its peak. Artificial Agency has raised<\/a> US$16 million in venture funding to date, from backers including Flying Fish, Toyota Ventures and Radical Ventures. Two of the startup\u2019s original co-founders left earlier this year, with one joining DeepMind Montreal.<\/p>\n \t\t \tAdam White, left, Martha White and Alden Christianson, not pictured, co-founded RL Core Technologies, a startup selling industrial automation software for water treatment plants. Photo: Larry Wong for The Logic\t<\/p>\n Edmonton remains \u201can underdog,\u201d compared to other tech clusters in Canada, according to Martha White, a UAlberta professor and CEO of RL Core Technologies. Despite that, or perhaps because of it, \u201cit\u2019s a very supportive ecosystem,\u201d she says. DeepMind\u2019s closure helped return some talent to the pool, and RL Core Technologies is among the beneficiaries\u2014her co-founders, husband Adam White and Alden Christianson, are both DeepMind Alberta alumni. Founded in October 2023, the firm uses<\/a> reinforcement learning to help automate industrial facilities like water treatment plants. It\u2019s an extension of work the Whites had been doing<\/a> at Amii before DeepMind quit Alberta.\u00a0<\/p>\n The technology saves clients money by reducing the amount of power and chemicals they use, while helping their filters and equipment last longer. \u201cWe\u2019re creating the software for industrial automation,\u201d says White, RL Core\u2019s CEO. The firm has raised US$5.79 million from backers including Flying Fish and TQ Ventures.\u00a0<\/p>\n Edmonton\u2019s AI startup scene is now in rude health. Linke-Boyko says the number of firms that pitch their ideas to Flying Fish has increased, and more students at the city\u2019s universities are considering entrepreneurship as an alternative career path to academia or a job at a Big Tech firm. \u201cThere\u2019s some really exciting momentum,\u201d Linke-Boyko says.<\/p>\n Post-DeepMind, it\u2019s also easier for Edmonton startups to access capital, according to Timbers, who advises RL Core backer TQ Ventures on AI investments. \u201cWe\u2019re now on the circuit of places that VCs visit,\u201d he says. Thanks in part to DeepMind, some of tech\u2019s most high-profile executives and investors \u201chave eaten at Tres Carnales and been to Oilers games.\u201d<\/p>\n In addition to his VC work, Timbers is lead research engineer for Ai2, a Seattle-based AI research non-profit, where he\u2019s applying reinforcement learning to LLMs. \u201cI want to work on the stuff that is most useful or most promising,\u201d he says; at the moment, that\u2019s generative AI. Sony AI, meanwhile, has snapped several former DeepMind staff to form a small Alberta research hub. Those jobs have kept people in Edmonton, where they\u2019re still collaborating today.<\/p>\n Alberta also continues to produce world-leading AI science. Sutton is currently working on step one of his 12-part Alberta Plan for AI Research, which he published<\/a> with Bowling and Pilarski shortly before DeepMind left Edmonton. Designed to last up to a decade, the plan tries to address what Sutton sees as a major problem with current machine learning systems like LLMs: they don\u2019t have goals, and they don\u2019t learn. \u201cAn essential part of intelligence is that you keep learning,\u201d he says.<\/p>\n The first step involves perfecting the basic units for a new kind of neural network that can generalize and learn continually. Sutton and DeepMind\u2019s Silver also recently published a paper predicting<\/a> that a new generation of AI agents will \u201cacquire superhuman capabilities by learning predominantly from experience.\u201d The article \u201chas caught people\u2019s imagination\u201d and increased interest in the Alberta Plan, Sutton says.<\/p>\n \t\t \tUniversity of Alberta professor Richard Sutton, left, is working on AI systems that learn continually, while Amii CEO Cam Linke helps researchers apply the technology to fields like medicine and space. Photo: Larry Wong for The Logic\t<\/p>\n Sutton has always focused on the fundamental, unanswered questions of AI, and he says DeepMind never pressured him to come up with commercial applications for his work. But he couldn\u2019t persuade the firm\u2019s senior leaders to focus on systems capable of continuous learning. It was also difficult to do \u201cscientific, systematic, careful, thoughtful research in AI\u201d at DeepMind, Sutton says, because the firm was too busy chasing \u201cthe hype and the buzz\u201d of the field. \u201cI think better outside of that company,\u201d says Sutton, who\u2019s wearing a DeepMind hoodie.<\/p>\n Sutton is doing his work on the Alberta Plan at UAlberta, Amii and at Keen Technologies, a startup he joined<\/a> in September 2023 that\u2019s pursuing artificial general intelligence. He\u2019s also helping raise money for the Openmind Research Institute, an Edmonton-based non-profit he co-founded<\/a> in November 2023 to back young AI researchers to pursue their scientific interests without the pressures of academia or industry.\u00a0<\/p>\n
\n\t![\"\"](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Richard_Sutton-Edmonton-Sept_2023-P_Jason_Franson-The-Logic-1920x1280--768x512.jpg\")
\t<\/a><\/p>\n
\n\t![\"\"](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Artificial_Agency_CEO_Brian_Tanner-co_founders_Alex_Kearney-Mike_Johanson-Andrew_Butcher-17123463767.jpeg\")
\t<\/a><\/p>\n
\n\tMurad Hemmadi and Jesse Snyder\n\t<\/p>\n![\"Three](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Lee_Se-Dol-Seoul-March_2016-P_Kim_Min-Hee-515358470-1920x1280-1.jpg\")
\t<\/p>\n![\"\"](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Alex_Kearney_Brian_Tanner-Edmonton-P_Larry_Wong-The_Logic-1920x1280-1.jpg\")
\t<\/p>\n![\"A](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Amii_logo_sign-Edmonton-May_2025-P_Artur_Widak-2213909739-1920x1280-1.jpg\"){kind=logo}
\t<\/p>\n![\"A](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Adam_White_Martha_White-Edmonton-2025-P_Larry_Wong-The_Logic-1920x1280-1.jpg\")
\t<\/p>\n![\"Two](\"https:\/\/www.newsbeep.com\/ca\/wp-content\/uploads\/2025\/09\/Richard_Sutton_and_Cam_Linke-Edmonton-2025-P_Larry_Wong_The_Logic-1920x1280-1.jpg\")
\t<\/p>\n