22 March 2021
While AI has come a long way since Alan Turing and other AI pioneers got the ball rolling, current AI is still a way off from truly intelligent machines. The gaming industry has jumped on recent AI advancements to create more realistic virtual worlds.
The convergence of blockchain, cryptocurrency and gaming industries has sparked a revolution toward transparent and open digital asset markets.
Artificial Intelligence is
playing a bigger role in the gaming industry than ever before.
Artificial intelligence (AI) refers to a group of concepts enabling computers to work like a human brain. For example, if you took millions of dog photos and fed them into an AIs neural network, it would recognise patterns and learn what a dog looks like, similar to the way children learn the words for animals. Defined in another way, artificial intelligence describes the broad field of intelligent machines performing tasks.
Neural networks are part of AI. They simulate the layered way the brain makes decisions but inside a computer, enabling it to learn, recognise patterns and make decisions. Machine learning is the part of AI permitting machines to learn by experience, acquiring skills without human interference.
While AI has come a long way since Alan Turing and other AI pioneers got the ball rolling, current AI is still a way off from truly intelligent machines. The gaming industry has jumped on recent AI advancements to create more realistic virtual worlds. Computer scientists have always considered games to be an effective way to measure an AIs strength. The rules are clearly defined and what constitutes success or failure is apparent.
Although AI is now widely used
to make modern games more realistic, the first AIs were built to play human
players at board games like checkers, chess and Go. Games have always provided a
concise way to measure an AIs strength as the rules are clearly defined, and
success is measurable. Computer scientists have been developing AIs for decades
to beat the best human players at our favourite games.
Arthur Samuel, famous for popularising the term "machine learning", developed a program called 'The Samuel Checkers Playing Program' in 1959, which, as the name suggests, played checkers. The program used a search tree of all possible board positions reachable from the current situation, as well as a scoring function to measure each side's winning chances.
After conquering checkers, chess was the next big challenge for AI pioneers. Although chess computers did exist in the early '70s, they weren't capable of beating any serious players. The first AI developed to beat top players was Deep Thought, which defeated a few grandmasters but lost its match against world champion Gary Kasparov 2-0.
The following version, called Deep Blue, played a match against Kasparov seven years later and also lost. Deep Blue received several significant upgrades in subsequent years, which led to a surprising victory in 1997 - the first time any AI defeated a sitting world champion at chess. Since the historic upset, AI developers have created more powerful AIs capable of beating human players at more complex games than chess. In 2016, Go world champion Lee Sedol was beaten by DeepMind's AI AlphaGo in front of 200 million live viewers.
Chess engines like Komodo use an algorithm called the Monte Carlo Search Tree (MCST), which considers every move in a given chess position, then looks at all possible responses and plays the move it deems most likely to succeed. All possible moves expand out like branches on a tree, with every layer of depth adding more branches to the tree. Algorithms like MCST help developers build powerful chess engines and AIs. However, they have less utility in modern video games, which have too many variables for a search tree to account for. They're mainly used in turn-based strategy games like Civilisation and for Chess programs.
Many organisations are dedicated
toward building more powerful AIs to compete against humans in a variety of
The British company DeepMind successfully developed an AI for a ‘capture the flag’ variant of Quake III, called FTW. Their recent success indicates that AI could one day fill empty player slots in modern multiplayer games, among other uses.
Online team games have always posed a challenge for AI developers as players always have individual and team goals that must be simultaneously achieved. A post-trial survey of DeepMind's FTW AI indicated human players often preferred playing with AI teammates, finding them more cooperative.
In 2017 a Californian company called OpenAI developed an AI to challenge the world's best Dota 2 players, called OpenAI Five. The program won its first game against a pro player in 2017, and by 2018 it could play as a 5 man team against human opponents.
In 2019 OpenAI Five won consecutive games against reigning Dota 2 champions OG. The AI was made available to play against in Dota 2's arena, in which 99.4% of human challengers were defeated.
In modern games, AI is mainly used to create more engaging, responsive and adaptive player environments. It's also used to construct programs capable of beating human players at recent games like Dota 2 and Starcraft, as well as board games like Chess and Go.
Modern AI allows developers to
create adaptable NPCs (non-player characters) capable of learning and adjusting
their behaviour patterns in reaction to a player's style, allowing far more
Developments in neural networks and machine learning allow developers to create fascinating and organic NPC storylines and behaviours. However, the technology is still quite new and most game studios still rely on comparatively simplistic algorithms to program their NPC's behaviour. NPCs are currently unable to respond to a player's fighting style or movement patterns, leading to repetitive boss fights and gameplay.
The Finite State Machine (FSM) algorithm is one such algorithm still widely used to program NPC behaviour, used in games like Battlefield and Call of Duty.
Developers often use FSM to program guard or enemy NPC behaviour. The NPCs will engage a player in a fight if they see them but roam around if they are out of sight. The drawback to this kind of two-state algorithm is its repetitiveness and predictability.
Using modern machine learning-based AI, game designers can build virtual worlds filled with more adaptable NPCs than possible using algorithms like FSM. Rockstar, for example, developed Red Dead Redemption 2 (RDR2) which won international acclaim.
RDR2 developers successfully created an environment in which AI agents make their own decisions (within specific parameters), leading to unique playthroughs and a riveting storyline. Several YouTubers have spent hours documenting the game's NPCs' rich and exciting lives, some of which you can see here.
One drawback to using more sophisticated AI in games is unexpected NPC behaviour. Random or unpredictable behaviour often results in unrealistic NPC actions, so NPCs will always need boundaries in which they can operate.
AI developers have proved they
can produce AIs capable of beating the best human players at modern, complex
games like Dota 2.
It could still be some time before we see AI capable of beating human players at games with trillions of variables like Magic the Gathering.
Still, judging by the recent success of games like Red Dead Redemption 2, it's likely the focus of AI development will be on generating more realistic NPCs for players to interact with, and worlds to get lost in.
If you're interested in earning money in gaming tournaments, you can register for the beta below.