How Creative AI Can Fasttrack Innovation
By: Sarah Hoffman | SEPTEMBER 2, 2021
As companies try to figure out how and when to bring employees back to the office, one often purported benefit of doing so is that the proximity to colleagues – and the chance for spontaneous meetings and conversations – spurs innovation. Others argue, however, that no evidence supports that contention, and instead office culture may actually hamper innovation because needing to be in a prescribed place at specific times excludes some people. 1 And in fact, technology and artificial intelligence have enabled us to move beyond in-person experiences in many ways − just consider how we shop for food and clothing, chose a movie to watch, and even how we date. So can AI enable us to innovate more effectively, with anyone, at any time, and in any location, thereby reducing the need to be in an office?
  • Facebook.
  • Twitter.
  • LinkedIn.
  • Print

Well, what we know to date is that when it comes to rote tasks, AI has made tremendous progress. As automation has taken on a growing range of routine work, the share of US jobs requiring decision-making rose from 6% in 1960 to 34% in 2018, with nearly half that growth happening after 2007.2 But innovation is a different story. Innovation is about ideating, counter-intuitive approaches, and reading markets in fresh ways to suss out new opportunities. It’s about being creative. And it just so happens that AI is creeping into a new world that many considered immune to AI – creativity.

What’s Behind the Rise of Artificial Creativity?

Recently, AI’s been trying to flex its right-brain capabilities. AI is being used to create art, write poems, and even generate music, and is doing so well that it’s often hard to reliably distinguish human from algorithmic creative content.3 That’s why Amsterdam’s Rijksmuseum used AI to recreate parts of Rembrandt’s “Night Watch” that were cut off when the artist finished the painting 70 years ago.4 Last year, the first International AI Song Contest was held, encouraging creative teams to experiment with AI and data as part of their songwriting process (check out the winning song here).5 And in March, digital art created by a robot named Sophia sold as a non-fungible token (NFT) for $700,000.6 Now she’s working on “Sophia Pop”, collaborating with musicians to write both music and lyrics.7

Artificial creativity is showing real promise because the technology is now:

Generative, not just predictive. In the past seven years, synthetically generated images of human faces have gone from rudimentary to truly life-like.8 And that just scratches the surface of what today’s AI can generate. An AI fed with pictures of more than 1,000 classical sculptures produced its own unique piece, Dio (which was actually made from the shredded remains of the computer that designed it).9 In July 2020, San Francisco–based start-up OpenAI released its new language-processing software, GPT-3, which writes poems, love songs, and even pickup lines on par with many humans.10 The text of a video game was recently generated by GPT-3 as well as a coding autocomplete tool, GitHub Copilot, that provides suggestions as you write code.11 China’s Wu Dao is 10 times bigger than GPT-3, trained using 1.75 trillion parameters compared to GPT-3’s 175 billion, and is multi-modal – it not only generates text but audio and images as well, plus it sings!12

Cheaper and more reusable. Due to the falling cost of cloud computing and progress in algorithm design, AI model training has become cheaper. It cost about $7 to train an image recognition system in 2020, down from about $2,000 three years earlier.13 Transfer learning – the process of creating new AI models by fine-tuning previously trained neural networks – has also made it cheaper and simpler to create AI models. Instead of training a neural network from scratch, developers can download a pretrained, open-source deep learning model and finetune it for their own purpose. The developer Arfa created fake My Little Pony characters by using transfer learning to fine-tune a model that generates fursonas, non-existent furry personas.14 After only 1 hour of training, Arfa’s model was already generating pretty convincing characters.15

Accessible to more than machine learning experts. It’s much easier for creative types to use AI today, even for those with limited coding abilities. Nvidia’s new Canvas tool lets an artist sketch simple shapes and lines with a palette of real-world materials like grass and clouds and then immediately fills the screen with photorealistic content.16 Founded in 2020, Loudly has built music-creation tools, training a Generative Adversarial Network (GAN) deep learning algorithm on eight million music tracks and 100,000 audio loops. Music like this hip-hop track can be created in seconds.

How Creative AI Can Help Drive Innovation

In a rapidly shifting market fueled by new competitors, technologies and consumer demands, companies need to be able to innovate quickly and continuously. Creative AI, AI that can generate content and ideas more easily and cheaply, can help us:

Address disruption. In the Disruption Dilemma, Joshua Gans discusses key challenges firms face in addressing disruption, including the ability to see a change coming and identify a response.17 Creative AI could help on both counts. Recently, an AI system trained on almost 40 years of science journals correctly identified 19 out of 20 research papers that have had the greatest scientific impact on biotechnology.18 In 2018, Warner Music Group acquired tech startup Sodatone which uses an algorithm to review social, streaming, and touring data to find promising unsigned talent.19 Similar techniques may help us see and prioritize upcoming disruptors, such as DeFi, quantum, shifting regulations, or even new expectations from Generation Z. AI could also help identify “responses”. As Nobel prize winner Linus Pauling said, “The best way to have a great idea is to have a lot of ideas,” and creative AI is well-suited to generating ideas. VisiBlends, a tool created by researchers at Columbia and Stanford, creates patterns that blend multiple images to guide brainstorming design sessions, producing 10 times as many creative results as unguided brainstorming sessions.20 For example, for the advertising concept “Starbucks is here for summer”, VisiBlends could help find numerous images related to “Starbucks” and “summer” and automatically combine them in a variety of creative ways. Ben Syverson, Design Lead at Ideo, challenged GPT-3 to generate ideas for helping young people save money and got back a bunch of solid ideas, including allowing users to share their saving goals with people who will cheer them on and challenge them, as well as showing users how much they could save if they splurged on new sneakers instead of taking an extra vacation.21

Create better products and experiences. There are many ways that creative AI could enhance products and experiences. Launched in October 2020, uses GPT-3 to generate text for blogs, products, and headlines based on user-provided word inputs.22 Given some impressive advancements in areas like music and art, Creative AI could also jumpstart new personalization initiatives. Founded in 2018, Endel creates personalized sound environments (soundscapes) unique to the user to improve productivity and sleep based on the time of day, weather, as well as a user’s heart rate and motion.23 Launched in July, Artifly (meant to evoke the phrase “Art on the Fly”) creates brand-new personalized art for users in less than a minute after users scroll through a selection of artwork and click the designs they like.24 This same technology that is enabling new individualized art and soundscapes to be created so quickly and effectively could also be used in other industries, really taking personalization to the next level.

Enhance innovation in a hybrid work environment. The world of academia offers a useful case study for how creativity can thrive among distributed teams. After rising for decades, by 2018 almost 80% of published US scientific articles were written by teams from multiple institutions.25 The same researchers also found that the unexpected loss of a colleague caused the same reduction to output whether or not coauthors were having regular in-person interactions.26 And this was before AI was heavily in use. To help drive distributed innovations, companies like Donut and Hallway are trying to recreate the workplace “water-cooler” moments virtually by connecting employees to others in the company that they don’t know or haven’t spoken to in a while.27 Similar to recommending a movie to watch or someone to date, AI could make workplace digital introductions pretty sophisticated, taking into account what employees are working on, their emails, and what they’re chatting about on collaboration apps. Add in creative AI – including idea generation and brainstorming with AI – and the physical office could become less essential for innovation.

Questions to Consider

Will artificial creativity lead to security and copyright risks? When Github Copilot launched, Matt Shumer, Co-Founder and CEO of OthersideAI, warned about the threat of blindly trusting code generated by Copilot after it put the database password directly in code.28 Are there other security risks to worry about as we turn to AI for more generative tasks? Beyond security risks, there are also copyright concerns. Some worry that Copilot could reproduce chunks of existing code and violate the rule that open-source code can’t be used for commercial purposes without proper licensing.29

What are the bias implications of creative AI? On one level, using creative AI to enhance distributed innovation efforts might curtail bias. As Dan Spaulding, Chief People Officer at Zillow, puts it, “How much creativity and innovation have been driven out of the office because you weren’t in the insider group, you weren’t listened to, you didn’t go to the same places as the people in positions of power were gathering?”30 How might integrating creative AI into the early phases of the innovation process help balance the pressures to work from an office and some of the biases that come with it? Could it be used to overcome not only geographic and language barriers, but also to enable a broader set of employees with different skillsets and in different roles to participate? Additionally, in traditional brainstorming sessions, ideas offered by senior leaders or more outspoken team members often carry more weight – could AI-driven idea generation counter some of this bias? Perhaps, but a word of caution: creative AI is still AI and therefore vulnerable to AI bias.

How can we prepare employees for the changes creative AI may bring? Creative AI is great at generating things but is not yet very good at identifying the best of what it generates. Given that, how should we expect creativity-oriented jobs and tasks to change? If there is a greater emphasis on sifting through results generated by creative AI rather than generating ideas from scratch, how should we train employees for this? How might design thinking change to incorporate creative AI?

While these are still early days of creative AI, using AI to enhance innovation efforts shows a lot of promise – especially in a time where flexibility is needed more than ever. During the pandemic, we’ve seen higher rates of women dropping out of the workforce.31 Perhaps using creative AI to allow for more distributed innovation will have an important side benefit: enabling companies to attract and retain a more diverse workforce. Now may be the perfect time to innovate how we innovate.

  • Facebook.
  • Twitter.
  • LinkedIn.
  • Print
3 Köbis, N., & Mossink, L. D. (2021). Artificial intelligence versus Maya Angelou: Experimental evidence that people cannot differentiate AI-generated from human-written poetry. Computers in human behavior, 114, 106553.
8 Zhang, D., Mishra, S., Brynjolfsson, E., Etchemendy, J., Ganguli, D., Grosz, B., ... & Perrault, R. (2021). The ai index 2021 annual report. arXiv preprint arXiv:2103.06312.
13 Zhang, D., Mishra, S., Brynjolfsson, E., Etchemendy, J., Ganguli, D., Grosz, B., ... & Perrault, R. (2021). The ai index 2021 annual report. arXiv preprint arXiv:2103.06312.
17 Gans, J. (2016). The disruption dilemma. MIT press.
18 Weis, J.W., Jacobson, J.M. (2021) Learning on knowledge graph dynamics provides an early warning of impactful research. Nat Biotechnol.
Chilton, L. B., Petridis, S., & Agrawala, M. (2019, May). VisiBlends: A flexible workflow for visual blends. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (pp. 1-14).
25 Clancy, Matthew, "The Case for Remote Work" (2020). Economics Working Papers: Department of Economics, Iowa State University. 20007
26 Clancy, Matthew, "The Case for Remote Work" (2020). Economics Working Papers: Department of Economics, Iowa State University. 20007
31 Thomas, R., Cooper, M., & Cardazone, G. (2020). Women in the Workplace 2020. McKinsey & Company and Lean In.
Please enter a valid e-mail address
Please enter a valid e-mail address
Important legal information about the e-mail you will be sending. By using this service, you agree to input your real e-mail address and only send it to people you know. It is a violation of law in some jurisdictions to falsely identify yourself in an e-mail. All information you provide will be used by Fidelity solely for the purpose of sending the e-mail on your behalf.The subject line of the e-mail you send will be " "

Your e-mail has been sent.

Your e-mail has been sent.