It is widely accepted that in the local universe stars form in giant molecular clouds (GMCs) where their optical and mostly UV light is strongly absorbed by the dust which surrounds them. Whether extinction was already taking place in the more distant universe where galaxies are less metal rich was less obvious ten years ago. Galaxies forming stars at a rate larger than about 20 M yr-1 were known to radiate the bulk of their luminosity above 5 µm thanks to IRAS, the so-called luminous (LIRGs, 12 > log(LIR / L) 11) and ultra-luminous (ULIRGs, log(LIR / L) 12) infrared (IR) galaxies. In the past, when galaxies were more gaseous and formed the bulk of their present-day stars, it would have been logical to expect to detect the past star formation events of galaxies in the IR regime and to detect a large population of LIRGs/ULIRGs too. However, prior to the launch of the Infrared Space Observatory (ISO, Kessler et al. 1996), this idea was not widely spread. Partly because of a cultural reason: star formation rates (SFR) were commonly measured from optical emission lines and rest-frame UV light in galaxies. This may explain why the redshift evolution of the SFR density per unit comoving volume computed by Madau et al. (1996) became famous. However in the first presentation of the density of UV light per unit comoving volume (Lilly et al. 1996) the authors were cautious to avoid converting their UV light into a SFR because of the unknown factor to correct for extinction. Already IRAS observations indicated a rapid decline of the comoving number density of ULIRGs since z ~ 0.3 (Kim & Sanders 1998, see also Oliver et al. 1996), but this was over a small redshift range and with small number statistics. In the few years that followed the launch of ISO, several observations showed that galaxy formation could not be understood, at least on an observational basis, without accounting for dust extinction as a major ingredient. The ISO deep surveys played a major role in this process, together with other results summarized below. They clearly established that extreme events such as those taking place in local LIRGs and ULIRGs must have been more common in the past, so much that they can now be considered as a standard phase that most galaxies experienced during their lifetime, at least once, but maybe even several times.
The first result of the ISOCAM surveys, as well as the ISOPHOT ones, was the great difference of the counts measured at faint flux densities with respect to local ones from IRAS (Elbaz et al. 1999, Dole et al. 2001). The universe must have been much richer in IR luminous galaxies in the past, either because galaxies were more IR luminous, at fixed galaxy density, and/or because the number density of galaxies was larger in the past, which was partly expected due to the reverse effect of hierarchical galaxy formation through mergers. The strength of the excess of faint objects came as a surprise, but its consequences on the past star formation history of galaxies was confirmed by the convergence of other observations going in the same direction:
- the nearly simultaneous discovery of the cosmic infrared background (CIRB, Puget et al. 1996, Fixsen et al. 1998, Hauser & Dwek 2001 and references therein), at least as strong as the UV-optical-near IR one, whereas local galaxies only radiate about 30% of their bolometric luminosity in the IR above ~ 5 µm .
- the 850 µm number counts from the SCUBA sub-millimeter bolometer array at the JCMT (Hughes et al. 1998, Barger et al. 1998, Smail et al. 2002, Chapman et al. 2003, and references therein) which also indicate a strong excess of faint objects in this wavelength range, implying that even at large redshifts dust emission must have been very large in at least the most active galaxies.
- the most distant galaxies, individually detected thanks to the photometric redshift technique using their Balmer or Lyman break signature showed the signature of a strong dust extinction. The so-called "-slope" technique (Meurer et al. 1999) used to derive the intrinsic luminosity of these galaxies and correct their UV luminosity by factors of a few (typically between 3 and 7, Steidel et al. 1999, Adelberger & Steidel 2000) was later on shown to even underestimate the SFR of LIRGs/ULIRGs (Goldader et al. 2002).
- the slope of the sub-mJy deep radio surveys (Haarsma et al. 2000).
It has now become clear that the cosmic history of star formation based on rest-frame UV or emission line indicators of star formation such as [0II] or [H] strongly underestimate the true activity of galaxies in the past if not corrected by strong factors due to dust extinction. Although distant galaxies were less metal rich and much younger, they must have found the time to produce dust rapidly in order to efficiently absorb the UV light of their young stars.
We have tried to summarize in the following the role played by the ISO deep surveys in establishing this new perspective (see also Genzel & Cesarsky 2000). However, ten years after ISO's launch we are still trying to understand the consequences of these findings on galaxy formation scenarios. Are these distant LIRGs really similar to local ones ? What do they teach us about the connection between star and galaxy formation on one side (IMF, triggering of star formation, conditions of star formation,...) and between galaxy and large-scale structure formation on the other side (role of the environment in triggering star formation events, galaxy versus group and cluster formation, ellipticals versus spirals, ...) ? How much energy radiated by compton thick embedded active nuclei remains undetected even by the Chandra and XMM-Newton X-ray observatories ? These questions together with others that will be discussed in the following demonstrate the liveliness of this field that will continue to feed the next generation of telescopes and instruments to come such as Herschel, ALMA, the James Webb Space Telescope (JWST) or the Spitzer and GALEX space observatories presently in use.