The artificial intelligence (AI) industry is undergoing unprecedented growth. AI development is finding its way into all areas of business and, increasingly, into creative endeavors. Despite the obvious difficulties, Canva, The Grid, and Autodesk have all tried to market AI design algorithms, with varying degrees of failure. Adobe is understood to be working on AI enhancements to its Creative Cloud product range. Eventually, someone will arrive at a mathematical formula that approximates the creative process.
Until recently, the emphasis of AI design has been on substituting an algorithm for a designer. Now, for the first time, a startup in Adelaide, Australia, is bringing to market an AI client.
Brigmore Technologies’ “Garvey” project is the culmination of 12 years’ research that began as a simple brief writing application and evolved into what is potentially an industry redefining service.
Project Garvey is an AI client designed to replace human clients in the design process. It works by scanning popular submissions on portfolio sites, including Dribbble, Behance, and DeviantArt— the team behind Garvey hopes that it will eventually have built a large enough dataset to scan its own records; from which it attempts to extrapolate a brief that could have prompted the work.
Think of it as a game of design Jeopardy.
For years designers have been validating decisions based on split testing; determining whether to use a green button or a red button based on the response each version gets. It’s a natural extension for an AI client to write a brief based on the parameters it finds elsewhere.
Once a brief is prepared, Garvey acquires designers by auto-submitting the project to major freelancing sites, like 99designs, fiverr, and freelancer. By version 1.3 of the AI client (the current version is 1.0.11) the team plans to have integrated a cold-contact chatbot, to enable Garvey to solicit work from designers and agencies directly.
Once commissioned, the designer can expect regular feedback from Garvey. At least once per day, the designer will receive a “How’s it going?” email or SMS—early trials with both Slack and Skype found designers easily identified Garvey as an AI when the conversation was too rapid.
To obfuscate the chatbot’s identity, and to ensure designers receive a full range of feedback, Garvey will adopt the identity of numerous fictional staff members. The CEO might chip in, as might the head of sales. As an added touch of realism, the opinions, and instructions from the “staff” will often be contradictory.
The Granddaughter Algorithm
According to Brigmore’s CTO Tom Mewling, the most challenging aspect of developing Garvey has been building the intelligence engine that responds critically to designs. The subjective, decision making algorithm of Garvey has been nicknamed “Granddaughter”, or GD for short. “If GD doesn’t like a design, then Garvey won’t sign-off,” explains Mewling.
The team set out to build in a degree of craziness into the GD algorithm, in order to emulate the nightmare clients we’ve all had, but it turns out that craziness was inherent: AI isn’t bound by the rules that humans don’t realise they’re following, meaning Garvey’s logic doesn’t always match a human’s, consequently Garvey is capable of some bat-shit craziness.
Whether Garvey’s more eccentric requests will make it past GD is another matter: “We’ve had instances in private beta, when Garvey explicitly demands a feature, the designer delivers, and then GD vetoes it ahead of the final sign-off,” says Mewling.
Each brief is a living, iterative process. As a designer delivers revisions, so the AI client adapts the brief.
Each brief is a living, iterative process. As a designer delivers revisions, so the AI client adapts the brief. Garvey has built into its logic an entitlement to unlimited revisions; in over 2,440 beta tests, no designer ever reached final sign-off.
There are legal ramifications to this issue: In the EU, new data protection rules the General Data Protection Regulation (GDPR) come into effect in 2018. GDPR means that within Europe, decisions cannot be based on data processing alone; the same protection does not exist in the US, Canada, Australia, or by 2019, in the UK.
Fixing the Client Problem
The ultimate goal of the Garvey project is to bring clients inline with the needs of designers. “Designers know far better than clients what gets comments on Dribbble, or is most likely to be nominated for an Awwward SOTD,” explains Mewling. “By removing human clients from the design process and replacing them with AI, we give designers the freedom they badly need. If designers can’t solve a client’s problem, then let’s rewrite the problem to fit the solution.”
The ambition of the team behind Garvey is to have completely replaced human client work by 2025.
That’s an ambitious goal. The growth of AI clients is inevitable, but comes with significant downsides. Briefs issued by Garvey for example, never offer financial compensation.
By removing human clients from the design process…we give designers the freedom they badly need
“It’s our experience that really committed designers don’t necessarily want to get paid,” says Hannah Grieg, Senior Marketing Executive for Brigmore Technologies. “What designers want is the opportunity of amazing exposure, and work that will look really good in their portfolio.”
While it’s certainly true that the vast majority of design is unpaid—designers spend time doodling in sketchbooks, posting on Medium, or working on personal projects—it’s not unreasonable to think designers may want more compensation than simple validation from a chatbot.
Despite projects commissioned by Garvey rarely, is ever reaching final sign-off, Brigmore Technologies does offer some hope of financial compensation. Those designers that do achieve sign-off will be approached by the AI for repeat business, and in future, once Garvey is profitable, there will be the opportunity of paying work.
Of course, not all AIs are created equal; the underlying datasets, codebases, and algorithms vary greatly.
Launching in Q3 2017, is Garvey Pro, a commercial version of the AI client. Garvey Pro will provide designers with the opportunity to purchase different service levels. Poor quality clients, with an uninspiring brief and poor attitude will be relatively affordable. High quality briefs will be restricted to enterprise level accounts.
This is intended to reflect the real-world spread of design projects, with the high-profile work being awarded to established agencies who can afford to pitch.
Will Garvey Redefine the Design Industry?
Everyone involved in the design process is arguably the designer. A good brief is a vital part of that process, and considered feedback guides a design throughout a project lifecycle.
Managing up to 164 million briefs at a time, with potential to scale, Garvey is wide ranging enough to keep the entire industry busy. Garvey is, at its core, a complex learning mechanism; as more designers are hired by the AI it will evolve until it is indistinguishable from a human client. The definition of designer’s educating their client.
Once AI design applications perfect their process, it’s entirely possible that a designer will be able to turn on their machine, launch their design app, and their client app, and let the two interact. That, according to Mewling, is what counts: “Most designers are just grateful to have client work in their portfolio, is shouldn’t matter if it came from a chatbot.”
Replacing human clients with an AI is an industry redefining moment, because for the first time, client briefs will be based on average design trends, instead of complex real-world problems.
Despite the obvious advantages of a design industry run by AI, there are still nods to human-clients. “Some of the feedback is pulled straight from classic human dialog,” says Mewling. “It doesn’t matter how large you start, Garvey will always ask you to make the logo bigger.”