Table 1. Work proportion and degree of division of labor obtained under different behavioral architectures.

		r = 0			r = 0.5
Work ratio	Result	c = 0	c = 1	c≥2	c = 0	c = 1	c≥2	Model
β  = 0.5	p 1 =  β	Y	Y	Y	Y	Y	Y	RT
		N	P	P	N	N	P	FFN
		Y	P	P	Y	Y	P	RN
	D >0.5	N	P	P	N	N	P	RT
		N	P	P	N	N	P	FFN
		N	P	P1	N	Y	Y	RN
β  = 0.75	p 1 =  β	N	N	N	N	N	N	RT
		Y	P	N	Y	P	N2	FFN
		Y	P	P3	Y	P	N3	RN
	D >0.5	N	N	P	N	N	P	RT
		N	N	P1	N	N	N	FFN
		N4	N	P3	N4	N	N3	RN

¹for c≥3.

²except small percentage (<3%) when c = 2.

³with exception of few simulations, where all colonies obtain the result.

⁴D<0.

Overview of results obtained for three different behavioral architectures: RT – response threshold model (A. Duarte, I. Pen, L. Keller and F.J. Weissing, subm.); FFN – feedforward neural network; RNN – recurrent neural network. Parameter combinations are indicated in the first column and first two rows. The second column indicates the result we look for: “” corresponds to the achievement of the optimal work ratio; “” corresponds to the evolution of worker specialization. In the central columns, for different levels of switching costs, c, we indicate if such results were obtained. “Y” indicates it was satisfied in all replicate simulations; “N” indicates that the result was not obtained, in the majority of simulations; “P” indicates that, in the majority of simulations, a fraction of the colonies within the population obtained the result.