(links and images still need to prettify’d, but the text is set)
https://commons.wikimedia.org/wiki/Robot#/media/File:Robot_DJ.jpg By Beaver, Brian (http://www.flickr.com/photos/bitboy/1398128600/in/photostream, CC BY 2.0, https://commons.wikimedia.org/w/index.php?curid=4258560) Artful Intelligentsia
I’ve beaten around the bush long enough, it’s time to get into some tech as it relates to beer. I’m going to go over some of the more interesting pieces of information I have come across lately. Each of these reflects some level of past, present, and/or future relationship we as consumers (and citizens) have with technology, and beer. In truth they are a snapshot of what is presently available and aspired to, containing a mixture of all three time perspectives.
True to the techie theme this edition will have more audio and video than the typical release. Enjoy the words and images in all their forms.
Apps Perhaps the simplest place to begin is with the availability and plethora of beer apps (https://beerconnoisseur.com/articles/app-takeover-beer-apps-ios-and-android) . This piece from Beer Connoisseur providing just a sample of what is possible. Find a beer, rate a beer, pair a beer. Not sure about the ABV? We got an App for that. There are game Apps, beer delivery Apps, and much more. Is there some blowhard yapping about beer styles and you have a sneaking suspicion they don’t know what they’re talking about? Call them out on that.
This is a particularly relevant (beer) technology sector dealing with the one true competitor to having a pint in hand. Though most of us would prefer a delicious drink there’s no denying that we spend a lot more time messing with our smart phones. It may be more socially acceptable (for the time being) to have the latter always at the ready but the former will almost certainly make you more socially agreeable.
Free Sticking with the App (adjacent) theme we turn to a piece after my own heart, free beer. Who wouldn’t be interested in free beer? Free good beer, of course. In my aging days I prefer to pass on a free Bud and pay for a quality ale instead. It’s not just me, is it? In any case, we have here a good premise: a bot that leverages social media data in order to recognize events offering free beer (http://austininno.streetwise.co/2017/01/11/how-to-find-free-beer-at-austin-events-ai-bots-will-lead-the-way/) . The company and its eponymous named service, Free Beer AI (https://freebeer.ai/) , appears to be primarily a text based supervised learning algorithm.
We will get to the AI part later, meanwhile the piece describes, and at least this writer’s experience can confirm, that the words “free beer” are not always so readily broadcast. In order to identify these opportunities the system identifies certain features/signals. Two examples likely to raise the probability are the mention of cocktails and or sponsors at an event. The system is interactive enough to allow its users to confirm the presence of free beer or lack thereof for past events, further helping improve the algorithm. The founder, Harlan Beverly, justifies his service (as if any justification were needed) by referring to it as “actually useful”, “help[ing] people get something”, and its potential as a tool to help recommend events of interest to its users as it learns more about their preferences.
I’m all for the endeavor but the “AI” moniker slapped onto the service is just a marketing ploy as far as I’m concerned. The future aspirations of the service as an events recommender gives it away. Additionally, the user feedback into the service suggests more algorithm than AI (https://www.theatlantic.com/technology/archive/2017/03/what-is-artificial-intelligence/518547/) . We are all familiar by now with recommender systems (from Amazon to Google, Pandora to YouTube), this is hardly AI in any meaningful sense. Of course I am as guilty as anyone else in not calling something AI the moment it actually works [link quote source].
Driving You wanna talk about something that looks more like AI? Check out this video of Otto (http://ot.to/) , a self driving truck on a beer run (http://www.denverpost.com/2016/10/25/self-driving-beer-truck-colorado/) . On October 20, 2016 Otto drove what amounted to 120 miles down I-25, Fort Collins, through Denver, to Colorado Springs “in what Anheuser-Busch says was the world’s first commercial shipment by an autonomous vehicle.” [Denver Post] The opening video sequence shows empty cans rattling through a canning line and the cargo trucks that are their first temporary home. The message is clear, AB InBev is gonna make and crate beer more efficiently than ever and at 1.2 M truckloads a year that’s a huge advantage.
So did AB just let loose a semi on the road? Not by themselves, this was an effort “engineered by A-B, Uber-owned self-driving truck maker Otto, and [Colorado] state transportation officials.” [Denver Post (http://www.denverpost.com/2017/01/04/ces-2017-galvanize-colorado-transportation-data-science/) ] Additionally, the truck was not unmanned. Though they never had to take the wheel there was a driver on board the whole time as a fail safe precaution. These are interesting and anxious times. Nearly two million people are employed as truck drivers in the United States (https://www.bls.gov/ooh/transportation-and-material-moving/heavy-and-tractor-trailer-truck-drivers.htm) . If Otto and similar autonomous vehicles come on board it will be an enormous economic disruptor. At the same time the motivations are clear: road safety, reduced emissions, and beer (even if it’s not a brew I’d prefer it at least pushes the technology further; you need the Big Boys for certain things).
At Home Sticking with disruption, while scaling back our computer overload suspicions, we have Geoffrey A. Fowler, Personal Technology Columnist for the Wall Street Journal, getting together with Seth Mansergh, San Francisco Homebrewers Guild President, to test and taste the Keurig (http://www.keurig.com/) of home brewing, PicoBrew (https://www.picobrew.com/) . Mansergh forewarns Fowler the first homebrew batch usually does not turn out well due to all of the variables involved.
|The video steps through the process, from unpacking and fitting the ingredients to adding the water, all in sped up video with background music (montage (https://youtu.be/vK4gv11PTI8) !). Brewing takes ~2.5 hours, with an one week fast fermentation. Check out C||NET for a review where they really put PicoBrew through the paces (https://www.cnet.com/news/picobrew-pico-progress-report/) .|
The video culminates with a third batch, the first two being misses due to an incorrect nozzle attachment (resulting in a software update to prevent future error, reminiscent of Tesla (https://www.wired.com/insights/2014/02/teslas-air-fix-best-example-yet-internet-things/) ) in one instance and excessive heat on the second go-round. Fowler cleverly mitigated the latter issue by putting the third fermenting keg in the WSJ server room which has obvious temperature control. Progress is made though Mansergh picks up on a sweetness finish he identifies as unfermented sugar. Even with fast fermentation the yeast could have used a little more time apparently.
PicoBrew is pushing their service quite strongly. They were present at the latest CES where they presented an upgrade on their features, allowing for beer customization (https://www.cnet.com/products/picobrew-freestyle/preview/) . I even recently heard an advertisement for their service on a CNN podcast.
Is this “instant” beer a cheating of the homebrewer DIY mentality? Perhaps. If you feel this strongly enough you may go with a different tech approach (BrewPi (http://www.brewpi.com) ) that allows for additional “freestyle” features. However, this option appears quite sophisticated and I would recommend people curb their enthusiasm and be honest with their capabilities before proceeding.
AI Again, Twice We round up this edition with two related “AI” takes on commercial brewing. First up is a brewery and we end with a marketing and consulting firm.
IntelligentX Brewing Company (http://intelligentx.ai/) plans to use “AI” to help develop better beer. Each of four base beer styles come with bottle labels listing a web link directing customers to a survey of tasting preferences. The idea is to use customer input and an algorithm to tweak the beers. There is a human brewer in the mix that serves as the “last line of defense if the AI goes rogue”; though according to the brewer the algorithm recipes are mostly spot on but occasionally random and even impossible. Hence the “last line of defense”.
I have struggled with this video ever since I saw it a couple of months back. I thought of using it as part of the kick off to the present cycle of newsletters but held off for a variety of reasons, all of which ultimately boiled down to my not being able to pinpoint and adequately express my discomfort. Primarily I’m turned off by the casual assertion that an reinforcement learning (RL) system would be sufficient on its own to devise the perfect beer.
Of course, that is not exactly what’s being claimed here. The video mentions brewers throughout history searching for the “perfect recipe” but in introducing the service/product, the cutely named ABI (like abbey, get it?) for Automated Brewing Intelligence, it’s more modestly described as an algorithm that takes customer feedback in order to make better beer; better, not perfect. In the end I may need only wrestle with my interpretation and hangups. I am a big fan of data, its need, and it’s potential. At the same time the data is but one ingredient in the recipe, a critical one no doubt but also insufficient on its own.
With apologies to The End of Theory (https://www.wired.com/2008/06/pb-theory/) view there is a need for understanding the problem; a framework of interpreting what is happening and to make predictions; some subject matter expertise; some justification for collecting the data and assuring the right things have been measured (https://mitpress.mit.edu/books/raw-data-oxymoron) ; and then there are the machine learning (ML) tools themselves. The latter can only do so much given how carefully or not the other steps were taken. I’m really going off the rails here again.
https://player.vimeo.com/video/172395607 IntelligentX Brewing Company (Vimeo) Cutting to the chase, the mention of AI generally and RL specifically stinks of a marketing ploy. Don’t get me wrong, I do not doubt for a moment they are implementing ML techniques but I’m skeptical about the rest. Again, is this AI in a strong significant sense? Probably not. As for the RL, you need a lot of data to help inform an algorithm of this sort. I am doubtful that the feedback provided by drinkers will provide sufficient input to meaningfully inform the AI what to do in such a large space of brewing possibilities (in all fairness IntelligentX does also mention a Bayesian approach). As one on-camera customer mentioned, “if you use an algorithm won’t it come out tasting middle of the road?”
And that’s just the beginning of the extendable technique to other products: coffee, chocolate, and even perfume
As a sort of counterpoint to the AI brewery we have an “AI” infused marketing and consultation company. For me this is the most informative of the highlighted pieces. Gastrograph (https://www.gastrograph.com/) founder, Jason Cohen (https://www.crunchbase.com/person/jason-cohen-8#/entity) , takes the time to drop a serious amount of knowledge about their work involving (primarily) beer, tea, and wine from a quality, process, and marketing perspective.
https://soundcloud.com/theaipodcast/ai-beer-gastrograph-jason-cohen The AI Podcast (SoundCloud) I highly recommend the above interview for takeaways and quotes such as the following, which I will allow to speak for themselves:
- Not bullish on electric tongue/nose technology; not going to be pouring the beer into a machine
- IF they worked, providing an accurate reading of the chemical composition of a product (e.g., Gas chromatography–mass spectrometry (GC-MS (https://en.wikipedia.org/wiki/Gas_chromatography%E2%80%93mass_spectrometry) ) or High-performance liquid chromatography (HPLC (https://en.wikipedia.org/wiki/High-performance_liquid_chromatography) )), they would still NOT predict flavor!
- “You cannot, right now, predict the flavor of a product from chemistry because [of] the number of interactions, the halo effect, the masking effects, the time/intensity effects are just exponential. Flavor is not a monotonic linear combination of normally distributed attributes that can be predicted from a chemical composition.”
- Can’t model flavor, then they can’t model preferences; and EVEN IF it could model flavor, what would be the feedback loop for calibrating/confirming preferences?
“The human palate has a million years of evolution to be perfectly predictive for what humans like today and in the future.”
- Personalize/precision beer/coffee/tea?
- No mass production in 25-100 years, sooner in the food & beverage industry.
- People like to consider themselves unique (tasting) snowflakes but that is not exactly true; people fall into o Demographic tasting populations o Tasting archetypes o (the reason you can have market winning products, e.g., Coca-Cola, Bud; but that doesn’t mean it’s the best product for you)