In a nutshell, generative design is a goal-driven approach to design that leverages automation so that designers and engineers can:
have better insight into their designs;
make faster, more informed design decisions; and
explore more options using the power of computers.
Better Outcomes and Insight
As the designer, you specify which outcomes you want to achieve for your design and how they are measured. With your guidance, the computer produces sets of optimal designs, along with the data used to prove which design performs best against your goals. By analyzing how the generated designs measure up against the set goals, you can gain valuable insight into which design aspects impact the outcome and how.
Maximization of active shared spaces - Mars Innovation District - The Living
Faster, More Informed Design Decisions
Generative design can help you find better designs for your project more quickly by leveraging what computers are good at: computation and repetition.
Computers can generate and evaluate a huge number of design variants in only a fraction of the time it would take an individual designer, allowing you to learn what does and doesn't work at an accelerated pace.
Above: Design options generated - Mars Innovation District - The Living
A Greater Variety of Options
With a generative design approach, the initial design parameters you input are used to generate your potential design solutions, with the only limitation being how much computer power and time you have.
For example, using traditional computational design techniques, it's feasible for you to explore ten variants (or more, perhaps). However, using generative design, an algorithm can generate thousands of variants in mere minutes.
Above: Design options generated - Bionic Partition for Airbus - The Living
A Collaborative Approach
The aim of a generative design approach is not to replace designers - it is to augment human capability with computation power.
To illustrate this, this about the fact that a good generative design process using several metrics to analyze designs in a study will rarely generate a single output. Instead, it will almost always generate a range of outputs that the designer can choose from, all of which will already have been objectively assessed and ranked. It doesn't choose for you.