My home city of Chicago is known as the birthplace of improvisational comedy. Over the years, everyone from John Belushi and Tina Fey to Bob “Better Call Saul” Odenkirk have performed on local stages, honing their wits and ability to adapt when scenes go off script. But while clubs like The Second City and Improv Olympic have become famous for producing television and movie stars, many of these comedy troupes also have successful side hustles in corporate training, helping organizations build their employees’ problem solving and collaboration skills.

In many ways, real life (and work) is more like an improv show than a movie. Your team plan a project for weeks only to have everything go “off script” due to circumstances beyond your control. Customers ask weird, unexpected questions during sales meetings, forcing salespeople to discard their carefully rehearsed presentations and adapt. The best solutions often come not from rigidly structured “innovation processes” but rather from teammates casually riffing on each other’s ideas, and following inspiration wherever it leads.

This also shows why traditional, scripted training content (manuals, e-learning, videos) can be poor preparation for real-world challenges. Even when e-learning tries to be “interactive”, it reduces complex situations to simplistic multiple choice questions which fail to capture the chaos of the workplace.

However, with AI, learning departments now have a technology that is capable of improvisation, capable of presenting learners with open-ended questions and adapting to the user’s unexpected responses. But are organizations ready for a learning technology that doesn’t always stick to the script?

Open Worlds: The Power of Unscripted Content

When Grand Theft Auto 3 became the top selling video game of 2001, it mainly gained notoriety for letting players assume the role of a sociopathic gangster. However, as a game, GTA3 fundamentally transformed the genre with its “open world” concept. Rather than forcing players down linear paths with finite solutions – like Super Mario Brothers – GTA allowed players to explore a living city and approach missions from literally any angle. Want to stride through the front door of a rival gang’s hideout, guns blazing? You can do that. Want to climb the fence and sneak through the rear window? You can do that, too.

Organizations have long applied the same principle to training people for high-stakes situations. Whether it’s the Red Cross rehearsing for a hurricane or a corporate cybersecurity team preparing for a breach, unscripted or semi-scripted simulation exercises are a highly effective form of training. And including an element of unpredictability can help learners overcome the human tendency to “freeze” when faced with unexpected, high-pressure situations.

Traditionally, these types of simulation exercises required expensive consultants to referee the scenario and sometimes even hired actors to play disaster victims or hospital patients (in fact, one of my former improv comedian friends now runs a company providing actors for medical and emergency responder training). However, with AI, we now have a way of delivering dynamic, adaptive, scenario-based learning inexpensively at scale, the same way e-learning provided an inexpensive alternative to classroom-based lecture/discussion training.

Open the player in full screen

The Straight Story: When Scripted Content is the Better Choice

Not long ago, our team demonstrated some prototype AI simulations for customer service and anti-money laundering training to a major bank. The end users and their managers loved the realism of the scenarios, however the bank’s compliance department nearly had a collective heart attack. “How can we approve the content if we never know exactly what it’s going to say?” they demanded.

Granted, one could argue that banks don’t know exactly what their human training facilitators will say in workshops (a point we gently raised), but there undeniably are situations where some level of scripting for training simulations isn’t just preferable, but required.  This includes:

• Legally mandated compliance training where organizations face serious consequences if regulations aren’t communicated to workers exactly as written.
• Highly specific training requirements that call for workers to rehearse a specific set of situations (for instance an aircraft performing a water landing)
• Zero-error tolerance situations in fields like neurosurgery or nuclear operations, where any deviation from protocol could have catastrophic consequences.

However, even then the choice between scripted and unscripted isn’t binary. AI can enhance even the strictest compliance training through hybrid approaches, such as pairing traditional scripted e-learning or videos with AI-powered simulation exercises and virtual coaches or providing an AI agent with a prewritten situation  at least the starting point is the same.

Freeform with a Framework: What we Can Learn from Improv Comedy

While some dramatic actors are known for going “off script” and improvising lines (see Robert Downey Jr., Marlon Brando and Meryl Streep), comedy is a field where improvisation is widely encouraged and celebrated.

Of course, “improv” comedy isn’t just about getting on stage and acting silly.  For example, the Upright Citizens comedy ensemble – whose members include American comedy stars like Amy Poehler, Donald Glover and Aubrey Plaza – is famous for bizarre scenes like Abraham Lincoln traveling through time and becoming a techno DJ at a dance club. However, no matter how goofy things get, UCB performances follow well-defined guidelines to ensure the narrative remains coherent and builds towards a satisfying comedic payoff by the end of the show.

Some of UCB’s principles include:

– Start with a “grounded”, everyday scene (e.g., two farmers are feeding chickens)
– Pay attention for anything unusual or funny that emerges naturally from the dialogue (e.g. one farmer notices the chickens seem to be pecking messages in Morse Code)
– Start building on the unusual or funny concept through logical escalation and repetition (e.g. the chickens are conspiring to take over the farm)
– No matter how strange things get, stay at least partly grounded in the established parameters of the scene and don’t radically diverge from the original concept (e.g. it wouldn’t be appropriate for an ensemble member to suddenly declare the chickens are aliens and that the farmers need to fly to the moon to stop them)
– If a show consists of multiple scenes, bring back characters and concepts from previous scenes (e.g. if the next scene involves workers at an advertising agency, perhaps they are discussing how to sell running shoes to the upwardly mobile chicken demographic)

While the Upright Citizens Brigade guidelines don’t translate directly to AI-generated training simulations, the basic idea of having the AI agent adhere to a loosely defined structure and guidelines is highly relevant.  For instance, when developing an AI-generated patient communications role play for doctors, our team gave the AI leeway to generate its own scenarios within the following rules:

– Start by grounding the scenario with a few fixed parameters (e.g. before the first line of dialogue, establish what kind of facility the scene is set in, the exact role and authority of the user, the patient’s personality, and their underlying condition)
– Allow the patient to react naturally to the user’s words and actions, but have the AI watch out for certain positive or negative behaviors (e.g. if the user is giving obviously bad medical advice, the patient should react with increasing skepticism)
– Let the conversation take its course, but never let the user change the basic facts (e.g. if the AI originally decided that the patient had acid reflux but the user declares “I believe you have angina”, the AI should not retroactively change the patient’s condition to match the user’s misdiagnosis)
– For more complex role plays and simulations, divide the narrative into stages with clear rules for when to end one stage and move on to the next (e.g. divide a cybersecurity simulation into distinct stages for each phase of an incident)

Open the player in full screen

Storytelling with Statistics: What We Can Learn from Dungeons & Dragons

If you ask most people about Dungeons & Dragons, they’ll probably say it’s a weird game that nerds play in basements, involving elves, trolls, and a lot of dice-rolling.  And while there’s some truth to that stereotype (speaking from personal experience), “D&D” is actually a collaborative, improvised storytelling game that offers some great insights for how to do semi-structured, simulation-based training.

For the uninitiated, a typical Dungeons & Dragons session has one person – the “Dungeon Master” – telling a fantasy-themed story, while the other players act out the roles of characters in the story (i.e. wizards, warriors, whatever.)   And while players are free to have their characters attempt any action they can imagine, whether or not they succeed depends on dice rolls and probability tables – so if you say “I’m going to pick up the elephant-sized monster with one hand and throw it over a cliff” you’re welcome to try, but will likely have a 1% chance of success.

When developing AI-powered role play simulations, our team follows a similar model.  The AI agent creates a narrative related to industrial equipment sales or emergency medicine, and the learner plays the role of an account rep or an emergency medical technician.  The learners are welcome to try saying or doing whatever they like (e.g. “I tell the customer about how much they can save on electricity with our new high-efficiency model…” or “I throw some water on the patient’s head to see if they wake up!”) but the AI agent checks a table with probabilities to determine how likely that action is to succeed – then another table to determine the consequences of success or failure (e.g. “The customer yawns and says ‘Our existing equipment is already pretty energy efficient.’”)

Conclusion

Just as improv teaches comedians to think on their feet, unscripted AI interactions can build learners’ ability to adapt and apply skills and concepts in the real world.

While corporate training has traditionally focused on compliance and adherence to procedure, the organizations that thrive won’t be those with the most rigid protocols or the most detailed manuals. They’ll be the ones whose people can face the unexpected with confidence, creativity, and competence.

Because, in a work environment where disruption is the only constant,  the ability to improvise isn’t just a nice-to-have: it’s critical for survival.