There are two ways to define $A_\infty$ structures. The elementary one seems rather intuitive when I think of higher homotopies. I.e. an $A_\infty$ structure on a $\mathbb{Z}$-graded vector $A$ space is a sequence of maps $(m_n)_{n\in\mathbb{N}}$, where $m_n:A^{\otimes n} \to A$ is of degree $n-2$, satisfying the $A_\infty$-relations: $$\sum_{r+s+t=n-1}(-1)^{r+st}m_n\circ(id^{\otimes r}\otimes m_s\otimes id^{\otimes s})=0.$$
This definition is then usually "restated in a more efficient way". For example, in Keller's introductory paper:
Let $(\bar{T}A,\Delta)$ denote the reduced tensor coalgebra on $A$, then an $A_\infty$-stuctures is a degree $1$ coderivation that squares to zero.
This re-definition is very mysterious to me. What was the original motivation? How would Stasheff(?) have come up with it?