This month, in his occasional sojourn into things which could be the next big things, he looks at what Facebook has been up to, game playing and changing AIs, and the ongoing battle for Net Neutrality.
Facebook has announced in a blog post on the 9th of August that they will soon be rolling out a new way to watch videos on their social network, titled Watch. Upping their game from the video posts you can already make on the platform, they’re expanding from videos just being in people’s feeds into a dedicated video area, available across mobile, desktop, and TV app.
Watch appears to be centralized around the idea of ‘Shows'; Shows are made up of episodes — live or recorded — and follow a theme or storyline. It looks like they intend to make the new platform kind of somewhere between Youtube and Netflix, with content creators making episodic series centred around a theme, which theoretically is a bit different from the all encompassing juggernaut of YouTube. Coupled with the 55% ad revenue share for partnered content creators, Facebook is definitely making a strong entry into the market for creators to carefully consider.
The blog post lists a whole bunch of features aimed at helping users find content they’ll enjoy, such as watchlists, categories, popular for ‘x’ reason sections, and so on. They’ve even written a section regarding what they believe to be the most suitable fitting content types for the platform, such as shows that engage fans or communities, live shows that connect directly with their fan bases, or that follow a narrative arc or use a consistent theme. And then there will be live events such as sports or major political events such as elections.
As you can clearly see, they’re targeting what they believe to be areas that will be the most successful on their platform, not least of which are the kinds of things that would bring in incredible amounts of advertising revenue, in particular those live events and shows like MLB games.
The rollout of Watch is initially to a limited group of users in the US with expansion plans for more soon. In closing, the blog post from Facebook was as ever ebullient. “We’re excited to see how creators and publishers use shows to connect with their fans and community. You can learn more about making shows on our Media blog.”
The team at OpenAI have created a bot for popular competitive eSport Dota 2, that has learned, completely on its own, how to play the game, and has upped its game so much that it has defeated the top Dota 2 professional players, with a special show match at the largest Dota 2 tournament of the year, The International, against arguably one of the best players in the whole world, Danylo “Dendi” Ishutin, a Ukrainian Dota 2 player for one of the world’s top teams, Natus Vincere (Na’Vi).
The video of the event, broadcast live from the stage at The International, begins with some introductory showmanship to hype the event. Then, they cue up a pre-recorded video showing off some of its capabilities by beating a quite a number of other top pro players. In the first game, Dendi plays how he normally would, only to find that bot already has him beat, even in a strategy known as Creep Blocking.
Creep Blocking, for the uninitiated, is the technique of physically blocking the wave of units that travels from your base into the lane to attack the enemy tower. The benefit to blocking your own friendly units or “creeps” is that you slow them down getting into the neutral middle area of the lane and end up having a terrain advantage with more safety.
Dendi was already surprised when he reached the middle of the lane to find that there were no enemy units coming. He was heard audibly on the footage to have said “’Bot, give me creeps!”, out of sheer disbelief that it could have blocked so significantly better, and that there was such a large gap between them. After this, there was a fairly lengthy back and forth exchange that lasted a few minutes, where both combatants traded blows, with the bot showing an extremely aggressive approach, even saying “Can you please stop bullying me!”. This then culminates with a damage trade that Dendi knew he couldn’t win, saying “I’m dead”, right before it happened and he ‘dies’ the first time.
The game is played to 2 kills, so he still has a chance of beating it in this game. Returning to the middle lane, then followed a little bit of ‘creep killing’ to gain some XP, before Dendi engaged the bot again, this time right at the lower part of terrain in the very middle – Dendi lost almost instantly, but it was a close finish.
During the interlude, a couple of the development team went on stage to show off the bot, and they go into detail about how the bot has learned to be this good. One of the team being interviewed explains why.
“So this bot is quite unlike any bot you’ve seen before. So, we’ve coached it to learn just from playing against itself, we didn’t hard-code in any strategy, we didn’t have it learn from human experts, just from the very beginning it just keeps playing against a copy of itself. It starts from complete randomness and then it makes very small improvements and eventually reaches the pro level.”
The host asked them to elaborate on the “complete randomness” that they refer to, only to learn in disbelief that the Bot learned from no grounding of the basics at all. The team member elaborates:
“It’s super fascinating to watch it train. When it starts out, it’s just completely random, so most of the games just end in one of the Shadow Fiends dying to the tower, not necessarily the mid tower, they just run around the map and then they fairly quickly figure out a better strategy, and that’s to stay in base.” The host then comically remarks “I’ve tried that one.”
“So then they spend a while basically AFK in the base, maybe occasionally they venture out, they hit a creep, they push out the wave a little bit and it helps a little bit, and maybe eventually they get to last hit.” At this point the host attempts to get to the heart of what huge leap lead to them going from this state to beating every pro player it comes up against. “It’s hard to say that there’s a particular big breakthrough, it’s a series of small improvements. They reach the level of casual players and they learn to play aggressively, then they learn to bait.”
The host then takes a crack at Dendi, saying “You know all about that don’t you Dendi?”. With a smirk, Dendi laughs, nods, and finally replies, “I know a strategy for my next game, like you said, AFK in base.” garnering a laugh from the audience.
Talking to the other member of the team on stage now, the host asks about how the bot has failed so many times and learned from it, to the point that it has played more than the pros. The team member responds “It’s played for really lifetimes of experience, it’s played so many games of Dota, it’s explored many different strategies, learned to bait, learned to exploit other people who bait, and it’s just played far more into the strategy space than any human has.”
Following this break was Game Two, where Dendi tells everyone his plan is to play more aggressively and push the lane, since it’s a different style of play to what he’s used to, being in a 1 versus 1 situation, as opposed to the game’s usual competitive 5 versus 5 mode. Then admitting that he’s going to try to copy the bot more to see how that works out, let’s just say it worked out just as expected: badly, as it was leaning already after how the first game went. This game ends even quicker, with Dendi proclaiming he’d already lost and leaving the game after the first death because of the advantage in dying, and not being in lane inherently – he concedes and acknowledges the victory of the bot.
Co-founded by Elon Musk in 2015, the closest thing we have to a real life Tony Stark, OpenAI is a non-profit research group with a full-time staff of 60 researchers and engineers that are developing “safe” AI systems with the goal of ensuring that “AI’s benefits are as widely and evenly distributed as possible.” Their goals being to benefit the world in any number of real world ‘messy’, imperfect situations to hear them describe it, such as the role of a surgeon.
The company is focused on long-term research into problems that require fundamental advances in AI capabilities. The group publishes at top machine learning conferences, releases blog posts to communicate its findings, and widely distribute open-source software tools for accelerating AI research.
As many of you may know, the fight for Net Neutrality, the principle that the internet should be an open place without Internet Service Providers (ISPs) having a stranglehold over giving websites preferential treatment, or slowing down traffic to competitor sites, or even blocking access to them altogether, in order to influence competition in the online space, was thought to have been won and finalised before Barack Obama stepped down from his presidency.
Since then, the question has be reignited and it seems the safety of the idea of an Open Internet is once again at stake. With this in mind, the Federal Communications Commission (FCC) has extended its comment period on the new proposal to the end of August, after advocacy organisations including the Electronic Frontier Foundation (EFF), Public Knowledge, and The American Civil Liberties Union (ACLU), filed a motion asking them for an extension in order to read through more comments.
The reason for such an extension is quite valid too, with some 20 million comments made. There is such a huge volume to read that it would definitely take time, with some of the comments more in-depth than others, as opposed to simple supporting or opposing statements. One such example was posted earlier this week from some of the representatives that were part of writing and modifying the Telecommunications Act itself. There have also been opposing views from several broadband and telecommunications groups, stating that there’s been ample time and it’s an old discussion.
The filing for extension of time also points out that there are a significant number of comments that appear fraudulent – from posters that have made duplicate comments, and those that have made comments from addresses that don’t exist, for example.
Whether this extension in time will be enough to mount enough of a case to help prevent the Open Internet we all know from an impending doom will remain to be seen.